<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Sarath SL on Medium]]></title>
        <description><![CDATA[Stories by Sarath SL on Medium]]></description>
        <link>https://medium.com/@slsarath2?source=rss-be7c33f2449b------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*CZNJ9y3ZTr1ZG4IJ2BYVuw.jpeg</url>
            <title>Stories by Sarath SL on Medium</title>
            <link>https://medium.com/@slsarath2?source=rss-be7c33f2449b------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 23:25:47 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@slsarath2/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[The New Four-Eyes Principle: Can One AI Agent Check Another?]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@slsarath2/the-new-four-eyes-principle-can-one-ai-agent-check-another-65bf8764490e?source=rss-be7c33f2449b------2"><img src="https://cdn-images-1.medium.com/max/1740/1*r0qfzlzb3Fp3ef_BSDD49A.png" width="1740"></a></p><p class="medium-feed-snippet">As banks move from AI assistants to autonomous agents, the traditional maker-checker model may need a rethink.</p><p class="medium-feed-link"><a href="https://medium.com/@slsarath2/the-new-four-eyes-principle-can-one-ai-agent-check-another-65bf8764490e?source=rss-be7c33f2449b------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://medium.com/@slsarath2/the-new-four-eyes-principle-can-one-ai-agent-check-another-65bf8764490e?source=rss-be7c33f2449b------2</link>
            <guid isPermaLink="false">https://medium.com/p/65bf8764490e</guid>
            <category><![CDATA[ai-governance]]></category>
            <category><![CDATA[risk-management]]></category>
            <category><![CDATA[agentic-ai]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[finance-and-banking]]></category>
            <dc:creator><![CDATA[Sarath SL]]></dc:creator>
            <pubDate>Wed, 07 Oct 2026 16:54:29 GMT</pubDate>
            <atom:updated>2026-10-07T16:59:13.895Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[FCA Priorities: What They Reveal About the Future of Proactive Risk Management]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@slsarath2/fca-priorities-what-they-reveal-about-the-future-of-proactive-risk-management-d5a2d246a436?source=rss-be7c33f2449b------2"><img src="https://cdn-images-1.medium.com/max/1536/1*_vWKXegzzBaTDTWZlgviUQ.png" width="1536"></a></p><p class="medium-feed-snippet">Regulation is no longer just asking, &#x201C;Do you comply?&#x201D; It is increasingly asking, &#x201C;Can you prove it with data?&#x201D;</p><p class="medium-feed-link"><a href="https://medium.com/@slsarath2/fca-priorities-what-they-reveal-about-the-future-of-proactive-risk-management-d5a2d246a436?source=rss-be7c33f2449b------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://medium.com/@slsarath2/fca-priorities-what-they-reveal-about-the-future-of-proactive-risk-management-d5a2d246a436?source=rss-be7c33f2449b------2</link>
            <guid isPermaLink="false">https://medium.com/p/d5a2d246a436</guid>
            <category><![CDATA[financial-regulation]]></category>
            <category><![CDATA[risk-management]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[financialconductauthority]]></category>
            <category><![CDATA[financial-institutions]]></category>
            <dc:creator><![CDATA[Sarath SL]]></dc:creator>
            <pubDate>Sun, 05 Jul 2026 15:36:05 GMT</pubDate>
            <atom:updated>2026-07-22T05:31:48.190Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[From Recovery to Anticipation: AI in Resilience Risk]]></title>
            <link>https://medium.com/@slsarath2/from-recovery-to-anticipation-ai-in-resilience-risk-42909453b810?source=rss-be7c33f2449b------2</link>
            <guid isPermaLink="false">https://medium.com/p/42909453b810</guid>
            <category><![CDATA[fintech]]></category>
            <category><![CDATA[resilience]]></category>
            <category><![CDATA[operationalriskmanagement]]></category>
            <category><![CDATA[genai]]></category>
            <category><![CDATA[ai-in-finance]]></category>
            <dc:creator><![CDATA[Sarath SL]]></dc:creator>
            <pubDate>Sun, 13 Jul 2025 17:21:52 GMT</pubDate>
            <atom:updated>2025-07-13T17:21:52.163Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8pixXcg3LZxJnJBikBhlUQ.png" /></figure><p>In today’s digital-first banking landscape, <em>resilience risk</em> is no longer just a line item in an audit report — it’s a boardroom concern. With increasing regulatory pressure, cyber threats, and technology dependencies, financial institutions are under immense pressure to ensure uninterrupted services and fast recovery from disruptions.</p><p>What if systems could <em>heal themselves</em>, proactively anticipate breakdowns, and even suggest solutions — without waiting for human intervention?</p><p>Thanks to Generative AI (GenAI) and Natural Language Processing (NLP), that future is already taking shape.</p><h3>🌩️ Understanding Resilience Risk in Banking</h3><p>Resilience risk refers to the <strong>threat of operational disruption</strong> that can impact an institution’s ability to deliver critical services. This includes:</p><ul><li><strong>IT system outages</strong></li><li><strong>Cybersecurity attacks</strong></li><li><strong>Third-party/vendor failures</strong></li><li><strong>Natural disasters or geopolitical events</strong></li><li><strong>Data corruption or loss</strong></li></ul><p>For banks, these events don’t just cause downtime — they risk <strong>regulatory non-compliance</strong>, <strong>financial loss</strong>, and <strong>reputational damage</strong>. With regulations like <strong>DORA (EU)</strong> and <strong>Basel III</strong> emphasizing operational continuity, banks must go beyond recovery — they must become <em>proactively resilient</em>.</p><h3>🛠️ What is a Self-Healing Risk Ecosystem?</h3><p>A <strong>self-healing ecosystem</strong> leverages automation and intelligence to <strong>detect</strong>, <strong>analyze</strong>, and <strong>respond</strong> to incidents with minimal human input. Think of it as an immune system for the bank’s operational landscape.</p><h3>Key Capabilities:</h3><ul><li><strong>Real-time anomaly detection</strong></li><li><strong>Automated incident triage and root cause analysis</strong></li><li><strong>Predictive alerts based on historical data</strong></li><li><strong>Dynamic recommendations for remediation</strong></li><li><strong>Continuous learning from past events</strong></li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ADk_I_oy5B99lbsSMcVFPw.png" /></figure><h3>🤖 Role of GenAI and NLP in Proactive Resilience</h3><p>While automation has existed for years, what’s new is how <strong>GenAI + NLP</strong> can interpret <strong>unstructured data</strong>, generate <strong>human-like insights</strong>, and simulate <strong>what-if</strong> resilience scenarios.</p><p>Here’s how it works:</p><h3>1. Incident Log Summarization</h3><p>NLP models can analyze logs from IT systems (e.g., Splunk, ELK) and automatically generate plain-English summaries of what went wrong — saving hours of manual RCA (Root Cause Analysis).</p><blockquote>Example:<em> Instead of digging through 15,000 log entries from a core banking failure, a GenAI-powered engine can summarize:</em></blockquote><blockquote><em>“Core banking module crashed at 03:42 AM due to memory overflow in payment gateway API triggered by an abnormal traffic spike.”</em></blockquote><h3>2. Auto-Generation of Remediation Playbooks</h3><p>GenAI can learn from previous incident resolution steps and suggest corrective actions in real-time — helpful for IT teams and crisis responders.</p><blockquote>Example:<em> During a recurring cloud outage, the system might recommend:</em></blockquote><blockquote><em>“Apply hotfix X1.04. Redirect traffic to backup server B. Notify vendor support and initiate Business Continuity Protocol 3.”</em></blockquote><h3>3. Simulating Disruption Scenarios</h3><p>Using historical data, GenAI can <strong>simulate synthetic resilience risks</strong> — like a data center failure or SWIFT outage — and evaluate the response time of various departments.</p><blockquote><em>This helps banks move from </em>reactive drills<em> to </em>AI-driven resilience testing<em>.</em></blockquote><h3>4. Real-Time Crisis Communication</h3><p>With NLP-driven chatbots or assistants, stakeholders (IT, Ops, Compliance) can receive real-time updates and FAQs during an incident.</p><blockquote><em>Imagine a chatbot embedded in MS Teams that answers:</em></blockquote><blockquote><em>“What systems are currently affected?”</em></blockquote><blockquote><em>“Has the regulator been notified?”</em></blockquote><blockquote><em>“What’s the estimated recovery time?”</em></blockquote><h3>🔎 Real-World Example: SWIFT Outage &amp; AI Opportunity</h3><p>In early 2024, a major European bank faced <strong>outbound payment failures</strong> due to a misconfiguration in its SWIFT messaging gateway. The issue remained unnoticed for <strong>6+ hours</strong>, impacting clients and requiring a regulator briefing.</p><p><strong>Where GenAI could have helped:</strong></p><ul><li>NLP-based alert system would have <strong>flagged unusual message queue build-up</strong></li><li>GenAI summarization would auto-generate an <strong>incident narrative for compliance</strong></li><li>Self-healing module could trigger <strong>fallback to secondary gateway</strong></li><li>Chatbot assistant would guide internal teams on <strong>next steps</strong></li></ul><p>The result? Faster detection, better coordination, less customer impact.</p><h3>🔧 Behind the Scenes (A Light Technical Peek)</h3><p>Though this blog is non-technical, here’s a high-level view of the architecture:</p><ul><li><strong>Data Sources:</strong> IT logs, alerts, service desk tickets, vendor feeds</li><li><strong>GenAI/NLP Engines:</strong> LLMs fine-tuned on incident narratives and compliance reports</li><li><strong>Integration Tools:</strong> APIs connecting to ServiceNow, Jira, Splunk</li><li><strong>UI Layer:</strong> Dashboards and chatbots for user interaction</li><li><strong>Feedback Loop:</strong> Continuous learning from new incidents and resolutions</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*eJhcWqcr7nC1lL6Fjs6TWw.png" /></figure><h3>📌 Final Thoughts</h3><p>Resilience is no longer about bouncing back — it’s about <strong>bouncing forward</strong>. As large banks continue navigating digital transformation and regulatory scrutiny, <strong>Generative AI and NLP offer a powerful shift — from reactive fixes to intelligent, automated response ecosystems</strong>.</p><p>By embracing self-healing architectures, financial institutions not only safeguard themselves from disruption but also position themselves as <em>trusted, future-ready institutions.</em></p><p><strong>Next Step for Banks:</strong></p><p>Begin with a <strong>pilot on post-incident RCA automation</strong> or <strong>GenAI-powered scenario testing</strong>. Partner with compliance and IT risk teams to define acceptable GenAI use cases, and build confidence gradually.</p><h3><em>Ready to turn your bank into a self-healing powerhouse? Start the conversation today.</em></h3><p>📌 <a href="https://www.linkedin.com/in/slsarath2/">Connect with me on LinkedIn</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=42909453b810" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[If a Model Decision Can Impact a Customer’s Life, It Must Be Explainable]]></title>
            <link>https://medium.com/@slsarath2/if-a-model-decision-can-impact-a-customers-life-it-must-be-explainable-2534f7199b47?source=rss-be7c33f2449b------2</link>
            <guid isPermaLink="false">https://medium.com/p/2534f7199b47</guid>
            <category><![CDATA[explainable-ai]]></category>
            <category><![CDATA[ai-regulation]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[compliance-management]]></category>
            <category><![CDATA[finance]]></category>
            <dc:creator><![CDATA[Sarath SL]]></dc:creator>
            <pubDate>Sat, 03 May 2025 12:48:36 GMT</pubDate>
            <atom:updated>2025-05-03T12:48:36.034Z</atom:updated>
            <content:encoded><![CDATA[<p><em>How Explainable AI (XAI) is shaping transparency in financial decisions.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*RuI4M_T5U1NLb26qjRrbTQ.png" /></figure><h3>Introduction: Why This Matters</h3><p>Imagine applying for a loan and getting rejected by an AI system. No reason, no explanation — just a “No.”</p><p>In finance, where decisions can affect a customer’s life, we need AI that not only works — but also explains itself. That’s where <strong>Explainable AI (XAI)</strong> comes in. It’s the bridge between smart machine learning and the human need for clarity, especially in areas like banking, credit, and fraud detection.</p><p>This blog will walk you through:</p><ul><li>Why explainability is critical in finance</li><li>What regulators expect</li><li>Techniques like SHAP, LIME, and counterfactuals</li><li>Real-world benefits for compliance, business teams, and customers</li></ul><h3>Where AI Is Used in Finance — And Why It’s Risky</h3><p>AI helps banks and financial institutions with big, repetitive tasks:</p><ul><li><strong>Fraud detection</strong>: Spotting unusual transactions in real-time</li><li><strong>AML (Anti-Money Laundering)</strong>: Flagging suspicious activities</li><li><strong>Credit scoring</strong>: Deciding if someone qualifies for a loan</li><li><strong>Insurance and claims</strong>: Predicting risk and fraud</li></ul><p>These systems bring speed and scale. But they also come with <strong>real consequences</strong>:</p><ul><li>A false fraud alert can freeze a customer’s account</li><li>A rejected loan can halt someone’s financial plans</li></ul><p>If these decisions are wrong — and we can’t explain them — that’s a big problem.</p><h3>The Black Box Problem: When You Can’t See Inside</h3><p>Many powerful AI models (like deep learning or ensembles) are <strong>black boxes</strong>. They make predictions, but even experts don’t fully know how.</p><p>Here’s why that’s risky:</p><ul><li><strong>No trust</strong>: Customers and managers won’t rely on a model they don’t understand</li><li><strong>Unfair outcomes</strong>: Models might unintentionally discriminate</li><li><strong>Regulatory issues</strong>: Laws require you to explain decisions</li><li><strong>Legal exposure</strong>: You may be sued or fined if you can’t justify a denial or alert</li></ul><blockquote><strong><em>Bottom line?</em></strong><em> If a model affects people — it must be explainable.</em></blockquote><h3>What the Rules Say: XAI Is Not Optional</h3><p>Global regulators are getting serious about transparency:</p><ul><li><strong>GDPR (EU)</strong>: People can challenge decisions made only by machines</li><li><strong>EU AI Act</strong>: High-risk AI (like in finance) must be explainable</li><li><strong>SR 11–7 (US)</strong>: Banks must manage model risk — including understanding how the model works</li><li><strong>CFPB (US)</strong>: You can’t use complexity as an excuse — lenders must give clear denial reasons</li></ul><p>So whether it’s Europe or the US, <strong>the message is the same</strong>: If your AI impacts people, you need to explain how it works.</p><h3>So What Is Explainable AI (XAI)?</h3><p><strong>XAI = tools and techniques that help us understand how AI models make decisions.</strong></p><p>Let’s break down a few popular ones:</p><h4>1. SHAP (SHapley Additive Explanations)</h4><ul><li>Think of each feature (like income, age, debt) as a player in a game</li><li>SHAP tells you how much each one contributed to the model’s final score</li><li>Great for local explanations (Why was this customer denied a loan?)</li></ul><h4>2. LIME (Local Interpretable Model-Agnostic Explanations)</h4><ul><li>Creates a small, simple model to mimic the complex one — just for one prediction</li><li>Gives a short list of reasons why the model decided what it did</li></ul><h4>3. Counterfactuals</h4><ul><li>Answers the question: <em>What would have changed the result?</em></li><li>Example: “If your income was $5,000 higher, your loan would have been approved”</li><li>Helps show fairness and provide actionable advice</li></ul><h3>How XAI Helps Real Teams</h3><p>XAI is not just for regulators — it helps everyone:</p><h4>✅ For Compliance Officers</h4><ul><li>Can clearly show why an alert was triggered</li><li>Easier audit trails and regulatory defence</li></ul><h4>✅ For Data Scientists</h4><ul><li>Debug models faster</li><li>Detect weird patterns or risky behaviour early</li></ul><h4>✅ For Business Leaders</h4><ul><li>Understand and trust model recommendations</li><li>Justify decisions to clients and regulators</li></ul><h4>✅ For Customers</h4><ul><li>Get clear reasons for adverse decisions</li><li>Feel respected and informed</li></ul><blockquote><strong><em>XAI builds trust — inside and outside the company.</em></strong></blockquote><h3>Example: AML Alerts With and Without XAI</h3><p><strong>Scenario</strong>: 500 alerts/day from a fraud detection model</p><ul><li><strong>Without XAI</strong>: Analysts don’t know why the alert was raised. They check everything from scratch.</li><li><strong>With XAI</strong>: Alerts come with reasons (“5 large transfers from high-risk countries”). Analysts triage faster.</li></ul><h3>Best Practices to Use XAI Well</h3><ol><li><strong>Design for explainability</strong> from day one</li><li><strong>Use simpler models</strong> when possible — or apply SHAP/LIME to complex ones</li><li><strong>Test and validate explanations</strong> (do they make sense to a human?)</li><li><strong>Tailor outputs</strong> for different users: analysts, business, customers</li><li><strong>Document everything</strong>: What features are used? What’s the model logic?</li><li><strong>Monitor drift</strong>: If the model’s behaviour changes, so should the explanations</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*YQXRk0NWtC2y7t91IzE0-A.png" /></figure><h3>Final Thoughts: Build AI That Can Be Trusted</h3><p>If AI is making decisions that impact people’s finances, it has to be:</p><ul><li><strong>Fair</strong></li><li><strong>Transparent</strong></li><li><strong>Understandable</strong></li></ul><p>Explainable AI makes that possible.</p><p>It’s not just about checking a compliance box — it’s about doing what’s right for the business, the regulators, and most importantly, the people.</p><blockquote><strong><em>A model that can’t explain itself is a model that may not belong in finance.</em></strong></blockquote><h3>📚 References &amp; Further Reading</h3><ul><li>EU AI Act: <a href="https://artificialintelligenceact.eu/">https://artificialintelligenceact.eu/</a></li><li>GDPR Article 22: <a href="https://gdpr-info.eu/art-22-gdpr/">https://gdpr-info.eu/art-22-gdpr/</a></li><li>CFPB Guidance on Adverse Action Notices: <a href="https://www.consumerfinance.gov/">https://www.consumerfinance.gov/</a></li><li>Federal Reserve SR 11–7 Guidance: <a href="https://www.federalreserve.gov/supervisionreg/srletters/sr1107.htm">https://www.federalreserve.gov/supervisionreg/srletters/sr1107.htm</a></li><li>SHAP Documentation: <a href="https://shap.readthedocs.io/">https://shap.readthedocs.io/</a></li><li>LIME Documentation: <a href="https://github.com/marcotcr/lime">https://github.com/marcotcr/lime</a></li><li>Counterfactuals in AI: <a href="https://arxiv.org/abs/1905.07697">https://arxiv.org/abs/1905.07697</a></li></ul><p><em>If you liked this post or have thoughts on how you use explainability in your work, drop a comment or connect with me. More insights coming soon on ethical AI in risk management.</em></p><h3>👤 Author Bio</h3><p>I’m a data scientist working in the financial sector, with a focus on control assurance, compliance analytics, and AI-driven automation. I’m passionate about making machine learning models more transparent and usable for real-world decision-making. I write to bridge the gap between technical complexity and business clarity.</p><p>📌 <a href="https://www.linkedin.com/in/slsarath2/">Connect with me on LinkedIn</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2534f7199b47" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Driving ESG Compliance Forward with NLP and Generative AI]]></title>
            <link>https://medium.com/@slsarath2/driving-esg-compliance-forward-with-nlp-and-generative-ai-da4abc142ec8?source=rss-be7c33f2449b------2</link>
            <guid isPermaLink="false">https://medium.com/p/da4abc142ec8</guid>
            <category><![CDATA[nlp]]></category>
            <category><![CDATA[esg-reporting]]></category>
            <category><![CDATA[finance]]></category>
            <category><![CDATA[banking-technology]]></category>
            <category><![CDATA[data-science]]></category>
            <dc:creator><![CDATA[Sarath SL]]></dc:creator>
            <pubDate>Sun, 12 Jan 2025 08:38:42 GMT</pubDate>
            <atom:updated>2025-01-12T08:38:42.029Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*6-6yGB2UIB7C0PgceXGKBA.png" /></figure><h4><strong>Introduction: The Growing Importance of ESG in Banking</strong></h4><p>In 2025, <strong>Environmental, Social, and Governance (ESG)</strong> compliance is not just a regulatory checkbox but a crucial part of sustainable finance. ESG refers to the three key factors used to evaluate the sustainability and ethical impact of investments in businesses. These factors include <strong>Environmental</strong> considerations such as carbon emissions and waste management, <strong>Social</strong> elements like employee diversity and community relations, and <strong>Governance</strong> aspects, which focus on corporate ethics and shareholder rights.</p><p>Banks play a pivotal role in driving ESG adoption by aligning their operations, investments, and risk frameworks with global regulations such as the <strong>EU Taxonomy for Sustainable Activities</strong>, <strong>Task Force on Climate-related Financial Disclosures (TCFD)</strong>, and <strong>Sustainable Finance Disclosure Regulation (SFDR)</strong>. Failure to comply with these standards can lead to heavy penalties, reputational damage, and loss of investor confidence.</p><p>However, the path to ESG compliance is fraught with challenges. Financial institutions must sift through enormous volumes of unstructured data, such as board reports, transaction logs, and third-party assessments, to identify ESG risks and adhere to reporting mandates. This is where technologies like <strong>Natural Language Processing (NLP)</strong> and <strong>Generative AI</strong> step in, offering scalable, efficient, and proactive solutions to streamline compliance processes.</p><p>This blog explores how NLP and Generative AI are reshaping ESG compliance in banking, with a focus on practical use cases and their relevance in today’s regulatory landscape.</p><blockquote><strong>The Domain Challenge: ESG Compliance in Financial Institutions</strong></blockquote><p>Banks grapple with several hurdles in achieving ESG compliance:</p><p>• <strong>Unstructured Data Overload:</strong> ESG risks are often hidden in textual data, such as regulatory filings, media reports, and legal agreements.<br>• <strong>Manual Effort:</strong> ESG compliance has traditionally relied on labor-intensive reviews, prone to human error.<br>• <strong>Global Standards:</strong> Compliance varies across regions, making it difficult to align with international ESG standards.<br>• <strong>Emerging Risks:</strong> Rapid changes in ESG requirements demand real-time adaptability.</p><p>Banks need innovative solutions to stay ahead, and that’s where NLP and Generative AI step in.</p><p><em>Resources:<br></em><a href="https://www.weforum.org/agenda/2023/03/what-is-esg-and-why-is-it-important-for-banks/"><em>What Is ESG and Why Is It Important for Banks? (World Economic Forum)</em></a><em><br></em><a href="https://www.fsb-tcfd.org/"><em>Task Force on Climate-related Financial Disclosures (TCFD)</em></a><em><br></em><a href="https://www.pwc.com/gx/en/services/sustainability/ai-esg-reporting.html"><em>The Role of Big Data and AI in ESG Reporting (PwC)</em></a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*0I0TlgQDN1koX1b7KTNEfQ.png" /></figure><h3><strong>NLP + Generative AI for ESG Compliance</strong></h3><ol><li><strong>Automating ESG Risk Detection</strong></li></ol><p>NLP helps banks automate the identification of ESG risks by analyzing unstructured textual data for specific patterns and keywords:</p><p>• <strong>Environmental Risks:</strong> Terms like “<em>carbon emissions,</em>” “<em>deforestation</em>,” or “<em>energy consumption</em>.”<br>• <strong>Social Risks:</strong> Indicators like “<em>gender disparity</em>,” “<em>workplace safety</em>,” or “<em>child labor.</em>”<br>• <strong>Governance Risks:</strong> Phrases signaling “<em>policy violations</em>,” “<em>bribery</em>,” or “<em>shareholder disputes.</em>”</p><p><strong>2. Named Entity Recognition (NER) for ESG Insights</strong></p><p>NER models extract meaningful entities from ESG documents, such as:</p><p>• Emission metrics: <em>“CO2: 500 tons”</em>.<br>• Legal entities: <em>“Company ABC violated X regulation”</em>.<br>• Financial impact: <em>“Fined $1 million for ESG non-compliance”</em>.</p><p><strong>3. Sentiment Analysis</strong></p><p>Sentiment analysis can process news articles or public sentiment around a company’s ESG initiatives, providing banks with early warnings for reputational risks.</p><h4>Example Code Snippet:</h4><pre>import spacy<br><br># Load NLP model<br>nlp = spacy.load(&quot;en_core_web_sm&quot;)<br><br># Example ESG-related text<br>text = &quot;Company X violated carbon emission limits, resulting in a $2M fine.&quot;<br>doc = nlp(text)<br><br># Extract key entities<br>for ent in doc.ents:<br>    print(f&quot;{ent.text} - {ent.label_}&quot;)</pre><p><strong>Generative AI: Adding a Proactive Layer to ESG Compliance</strong></p><p>While NLP enables banks to analyze existing data, <strong>Generative AI</strong> adds a new dimension: proactive content generation, summarization, and scenario analysis.</p><p><strong>1. ESG Report Summarization</strong></p><p>Generative AI models like ChatGPT can summarize lengthy ESG reports into concise, actionable insights for decision-makers. For example:</p><p>• Summarizing vendor ESG compliance reports.</p><p>• Highlighting key risks from annual board reports.</p><p><strong>2. Automating ESG Policy Drafting</strong></p><p>Banks must often draft ESG policies tailored to evolving regulations. Generative AI simplifies this by:</p><p>• Generating initial drafts of policies based on specific ESG requirements.</p><p>• Tailoring policies to meet regional regulations or industry standards.</p><p><strong>3. Enhanced Pattern Matching with Large Language Models</strong></p><p>Generative AI models can process and compare vast amounts of ESG-related data to detect discrepancies or inconsistencies:</p><p>• Comparing ESG narratives in vendor contracts with actual practices.</p><p>• Flagging potential ESG violations in news articles or public disclosures.</p><p><strong>4. Scenario Analysis</strong></p><p>Generative AI can simulate ESG scenarios to help banks anticipate and prepare for potential risks. For example:</p><ul><li><em>“What if CO2 emissions increase by 20% in the next quarter?”</em></li></ul><h3><strong>Why This Matters in 2025</strong></h3><p><strong>1. ESG Reporting is No Longer Optional</strong></p><p>With mandatory ESG disclosures now enforced in many countries, financial institutions must adopt cutting-edge technology to stay compliant.</p><p><strong>2. Scalability and Cost Efficiency</strong></p><p>The combination of NLP and Generative AI allows banks to process vast amounts of data across multiple formats and languages without additional resources.</p><p><strong>3. Competitive Edge</strong></p><p>Banks leveraging these technologies can offer better transparency to stakeholders, enhancing their reputation in a sustainability-focused world.</p><p><strong>4. Real-Time Adaptability</strong></p><p>Generative AI’s ability to generate and simulate future ESG scenarios empowers banks to stay agile in the face of evolving regulations.</p><h3><strong>Technical Challenges and Mitigation</strong></h3><figure><img alt="Image from Google" src="https://cdn-images-1.medium.com/max/710/1*X0blknSmluCcmphWCBfLLw.png" /><figcaption>Image <a href="https://www.pullrequest.com/blog/using-code-challenges-for-interviews/">from</a> Google</figcaption></figure><p><strong>Challenges:</strong></p><p>• <strong>Data Quality Issues:</strong> Unstructured or incomplete ESG data can hinder NLP accuracy.<br>• <strong>Bias in Generative AI:</strong> Generative models might produce biased summaries if training data lacks diversity.<br>• <strong>Integration Complexity:</strong> Merging NLP and Generative AI with existing systems can be challenging.</p><p><strong>Solutions:</strong></p><ul><li>Use advanced pre-processing techniques to clean data.<br>• Continuously fine-tune models with diverse and representative datasets.<br>• Work closely with IT teams to ensure seamless integration.</li></ul><h3><strong>Conclusion</strong></h3><p>The fusion of NLP and Generative AI is a game-changer for ESG compliance in banking. These technologies not only enhance efficiency but also enable proactive risk management, aligning with the global push for sustainability. As we move forward, the integration of these tools will define the future of ESG compliance, helping financial institutions navigate complexities with precision and foresight.</p><h4><strong>Call to Action:</strong></h4><p>How is your organization leveraging NLP and Generative AI for ESG compliance? Share your experiences or connect to explore how these technologies can be tailored to your needs.</p><p><a href="https://www.linkedin.com/in/slsarath2/"><em>https://www.linkedin.com/in/slsarath2/</em></a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=da4abc142ec8" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Simplifying Risk and Compliance in Finance with Generative AI: A Techno-Functional Approach]]></title>
            <link>https://medium.com/@slsarath2/simplifying-risk-and-compliance-in-finance-with-generative-ai-a-techno-functional-approach-087926ee9673?source=rss-be7c33f2449b------2</link>
            <guid isPermaLink="false">https://medium.com/p/087926ee9673</guid>
            <category><![CDATA[riskandcompliances]]></category>
            <category><![CDATA[risk-assurance]]></category>
            <category><![CDATA[gen-ai-revolution]]></category>
            <category><![CDATA[risk-management]]></category>
            <category><![CDATA[data-science]]></category>
            <dc:creator><![CDATA[Sarath SL]]></dc:creator>
            <pubDate>Fri, 17 May 2024 22:11:39 GMT</pubDate>
            <atom:updated>2024-05-17T22:15:46.644Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*bSEDMG5z2D9ulk2hNjA27g.jpeg" /><figcaption>Integration of Advanced AI Technology with Financial Data and Compliance Tools</figcaption></figure><p>In the dynamic landscape of financial services, managing risk and ensuring compliance are paramount. Traditional methods often struggle to keep up with the increasing complexity and volume of data. Generative AI offers a revolutionary approach, blending advanced technology with functional improvements to enhance risk and compliance processes. <strong><em>This blog explores how generative AI can be effectively utilised in financial institutions from both technical and functional perspectives.</em></strong></p><h4>Understanding Risk and Compliance</h4><p>Risk management involves identifying, assessing, and mitigating potential threats to an organization’s assets and earnings. Compliance ensures adherence to laws, regulations, and internal policies. Traditionally, these processes have been labor-intensive and prone to human error, presenting significant challenges in terms of efficiency and accuracy. The need for innovative solutions is more pressing than ever.</p><h4>Generative AI: A Brief Overview</h4><p>Generative AI refers to models that can generate new data similar to the input data they were trained on. This technology has shown remarkable success in various fields, from creating realistic images to generating coherent text. In the context of risk and compliance, generative AI can offer unprecedented capabilities in anomaly detection, predictive analytics, and process automation.</p><p>A recent advancement in generative AI is the introduction of ChatGPT-4o, a new model that has shown significant improvements over previous versions. ChatGPT-4o offers enhanced accuracy and efficiency, making it a powerful tool for financial institutions to manage risk and ensure compliance. This model can process large volumes of data more quickly and accurately, identifying potential risks and automating compliance tasks with greater precision.</p><h3><strong><em>Techno-Functional Applications of Generative AI in Risk and Compliance</em></strong></h3><h4>1. Anomaly Detection</h4><p>**<strong>Technical</strong>**: Generative AI models, such as autoencoders, can learn normal transaction patterns and detect anomalies by flagging deviations from these patterns. These models are trained on historical data to understand what constitutes ‘normal’ behavior.</p><p>**<strong>Functional</strong>**: By detecting anomalies, financial institutions can quickly identify potential fraudulent activities, enhancing security and operational efficiency.</p><h4>2. Predictive Analytics</h4><p>*<strong>Technical</strong>**: Machine learning models, including generative ones, can be trained on historical data to predict potential risks. These models analyze patterns and correlations within the data to foresee possible risk scenarios.</p><p>**<strong>Functional</strong>**: Predictive analytics enable proactive risk management. Financial institutions can take preventive measures before risks materialize, improving decision-making and reducing potential losses.</p><h4>3. Process Automation</h4><p>**<strong>Technical</strong>**: Generative AI can automate compliance-related processes by continuously monitoring regulatory changes and ensuring adherence to compliance requirements. Natural Language Processing (NLP) models can be used to parse and understand regulatory documents.</p><p>**<strong>Functional</strong>**: Automation reduces manual effort and ensures that compliance is maintained with minimal human intervention. This leads to significant cost savings and more reliable compliance processes.</p><h3>Case Study</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*K8RxBusjE-Ra39ZKjpLClw.jpeg" /></figure><p><em>Imagine a financial institution implementing generative AI to enhance its control testing tool. The objective is to improve accuracy, efficiency, and proactive risk management. Here’s a step-by-step breakdown:</em></p><p><strong>Technical Steps:</strong></p><blockquote>Data Collection: Gather PRO mapping data, including process IDs, risk IDs, and control IDs.</blockquote><blockquote>Model Training: Train a generative model (e.g., Random Forest or Autoencoder) using this data to learn patterns and predict potential risks.</blockquote><blockquote>Prediction: Use the trained model to predict risks in new transaction data.</blockquote><blockquote>Reporting: Generate detailed reports highlighting potential risk areas and control effectiveness.</blockquote><p><strong>Functional Outcomes:</strong></p><blockquote>The implementation leads to more accurate risk detection, reduced manual effort, and a proactive approach to risk management. Financial institutions can focus resources on high-risk areas, enhancing overall efficiency and security.</blockquote><h3>Advantages of Generative AI in Risk and Compliance</h3><ul><li>*<strong>Technical</strong>**: Generative AI offers improved detection rates for anomalies, scalability to handle large datasets, and increased model accuracy through continuous learning.<br>**<strong>Functional</strong>**: Financial institutions benefit from cost savings due to reduced manual effort, enhanced efficiency in compliance processes, and improved risk management by identifying and addressing potential issues proactively.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*OzhJmY8IjvDGVugRPJHmrA.jpeg" /></figure><h3>Challenges and Considerations</h3><ul><li>*<strong>Technical</strong>**: Ensuring data privacy, addressing model bias, and maintaining model robustness are key challenges. Implementing robust data governance frameworks and conducting regular model audits can mitigate these issues.<br>**<strong>Functional</strong>**: Adhering to regulatory compliance and ethical considerations is crucial. Financial institutions must ensure that their use of AI aligns with regulatory standards and ethical practices.</li></ul><h3>Future Trends and Innovations</h3><p>Emerging technologies like blockchain and quantum computing hold promise for further enhancing risk and compliance processes. Blockchain can provide secure and transparent transaction records, while quantum computing can offer unprecedented computational power for complex risk assessments.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-5NyShs-PPjmoZeuZzz_ag.jpeg" /></figure><h3>Conclusion</h3><p>Generative AI has the potential to revolutionize risk and compliance in financial institutions. By leveraging its capabilities, organizations can achieve greater accuracy, efficiency, and proactive risk management. As the technology continues to evolve, its integration into financial processes will become increasingly essential. Financial institutions should consider adopting generative AI solutions to stay ahead in this rapidly changing landscape.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=087926ee9673" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Unlocking the Power of Machine Learning in Transaction Monitoring for Anti-Money Laundering]]></title>
            <link>https://medium.com/@slsarath2/unlocking-the-power-of-machine-learning-in-transaction-monitoring-for-anti-money-laundering-f1c61305cd?source=rss-be7c33f2449b------2</link>
            <guid isPermaLink="false">https://medium.com/p/f1c61305cd</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[compliance]]></category>
            <category><![CDATA[anti-money-laundering]]></category>
            <category><![CDATA[transaction-monitoring]]></category>
            <category><![CDATA[banking]]></category>
            <dc:creator><![CDATA[Sarath SL]]></dc:creator>
            <pubDate>Tue, 07 Feb 2023 17:35:12 GMT</pubDate>
            <atom:updated>2023-02-07T19:32:15.236Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tIemQT6UHIBB1_J7-IjB-Q.jpeg" /></figure><p>The fight against money laundering and other illegal financial activities is an ongoing battle for financial institutions and government agencies. With the increasing volume of financial transactions, it has become increasingly difficult to manually monitor and detect suspicious activities. This is where the power of machine learning (ML) can help. By using ML algorithms and advanced analytics, organizations can quickly and accurately identify potential money laundering activities in real-time, reducing the risk of financial losses and ensuring compliance with regulatory requirements. In this blog, we will explore the benefits and implementation of ML in transaction monitoring for anti-money laundering (AML).</p><p><strong><em>There are several things that you can do using machine learning and deep learning for transaction monitoring to identify money laundering:</em></strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*IFxLrrXVbEYCnjIghM4mXw.png" /></figure><h3><strong><em>Anomaly detection</em></strong></h3><p>One of the simplest ways to identify money laundering activities using machine learning is by detecting anomalies in financial transactions. Machine learning algorithms like decision trees, random forests, and k-nearest neighbors can be used to identify transactions that deviate from the normal patterns. For example, a sudden increase in the frequency or amount of transactions can be considered an anomaly and flagged for further investigation.</p><h3><strong>Link analysis</strong></h3><p>Another way to identify money laundering activities is by analyzing the relationships between transactions, accounts, and individuals. Graph-based algorithms can be used to visualize these relationships and identify any suspicious patterns. For example, transactions that are frequently executed between the same individuals or between the same accounts can be flagged for further investigation.</p><h3><strong>Clustering</strong></h3><p>Clustering algorithms can be used to group similar transactions together and identify clusters of suspicious activity. Algorithms like k-means, hierarchical clustering, and Gaussian mixture models can be used to find transactions that have similar characteristics, such as transaction amount, merchant name, and transaction type. Once these clusters are identified, they can be further investigated to see if they contain any suspicious activity.</p><h3><strong>Fraud detection</strong></h3><p>Machine learning algorithms can be used to detect fraudulent transactions by building a model to distinguish between legitimate and illegal transactions based on past behavior and patterns. This model can then be used to predict future fraudulent transactions based on the historical data. For example, if a transaction is executed from an account that has never been used before, or if a transaction is executed from an account that has a history of fraudulent activities, it can be flagged for further investigation.</p><h3><strong>Predictive modeling</strong></h3><p>Deep learning algorithms like recurrent neural networks (RNNs) and long short-term memory (LSTM) networks can be used to predict future fraudulent transactions based on historical data. By training the model on large amounts of historical transaction data, it can learn to identify patterns and anomalies that are indicative of money laundering activities.</p><h3><strong>Unsupervised learning</strong></h3><p>Unsupervised learning algorithms like autoencoders and variational autoencoders can be used to identify and cluster suspicious transactions, even if there is no prior labeled data available. By using unsupervised learning techniques, you can find transactions that deviate from the normal patterns and investigate them further to see if they contain any suspicious activity.</p><h3><strong>Natural language processing</strong></h3><p>Finally, natural language processing (NLP) techniques can be used to extract relevant information from transaction descriptions and flag transactions that contain suspicious terms or patterns. For example, if a transaction description contains the term “money laundering,” it can be flagged for further investigation.</p><h4><em>In conclusion, the use of machine learning algorithms for transaction monitoring can greatly improve the accuracy and speed of identifying money laundering activities. By combining multiple techniques, including anomaly detection, link analysis, clustering, fraud detection, predictive modeling, unsupervised learning, and NLP, you can build a comprehensive solution for monitoring financial transactions and detecting money laundering activities.</em></h4><pre>LETS CODE :)</pre><p>Here is a simple example in Python that implements some of the techniques mentioned for detecting money laundering activities in financial transactions:</p><pre>import pandas as pd<br>from sklearn.ensemble import RandomForestClassifier<br>from sklearn.cluster import KMeans<br>from sklearn.preprocessing import StandardScaler<br><br># Load the transaction data into a pandas dataframe<br>df = pd.read_csv(&quot;transactions.csv&quot;)<br><br># Use StandardScaler to normalize the transaction amount column<br>scaler = StandardScaler()<br>df[&quot;amount&quot;] = scaler.fit_transform(df[[&quot;amount&quot;]])<br><br># Use KMeans to cluster the transactions into two groups<br>kmeans = KMeans(n_clusters=2)<br>df[&quot;cluster&quot;] = kmeans.fit_predict(df[[&quot;amount&quot;]])<br><br># Use RandomForestClassifier to build a fraud detection model<br>model = RandomForestClassifier()<br>model.fit(df[[&quot;amount&quot;, &quot;cluster&quot;]], df[&quot;fraud&quot;])<br><br># Predict the fraud probability for new transactions<br>new_transactions = [[1.5, 0], [0.5, 1], [-0.5, 1]]<br>print(model.predict_proba(new_transactions))</pre><p><em>In this example, we first load the transaction data into a pandas dataframe. We then normalize the transaction amount column using StandardScaler to ensure that the amount values are in the same range. After that, we use KMeans to cluster the transactions into two groups based on the amount values. Finally, we use the RandomForestClassifier to build a fraud detection model that uses both the transaction amount and cluster as features. Finally, we make predictions on new transactions by calling the predict_proba method of the model.</em></p><p><em>Please note that this is just a simple example to demonstrate the use of machine learning algorithms for transaction monitoring. In a real-world scenario, the solution would be much more complex and may involve multiple algorithms, data preprocessing techniques, and feature engineering.</em></p><h3>Step-by-step implementation</h3><ol><li><strong>transactions.csv</strong><br>This file contains transactions data, with columns for id, amount, type, and fraud. The id column represents a unique identifier for each transaction, the amount column represents the amount of the transaction, the type column represents the type of the transaction (debit or credit), and the fraud column represents whether the transaction is a fraud or not (1 means fraud, 0 means not fraud). This data can be used to train a machine learning model to identify money laundering activities.</li></ol><pre>id,amount,type,fraud<br>1,100,debit,0<br>2,500,credit,1<br>3,200,debit,0<br>4,400,credit,0<br>5,300,debit,1<br>6,600,credit,0<br>7,700,debit,0<br>8,800,credit,1<br>9,900,debit,0<br>10,1000,credit,0</pre><p><strong>2. Import the required libraries:</strong></p><pre>import pandas as pd<br>import numpy as np<br>from sklearn.ensemble import RandomForestClassifier<br>from sklearn.model_selection import train_test_split<br>from sklearn.metrics import classification_report, confusion_matrix</pre><p><strong>3. Load the transaction data into a pandas dataframe:</strong></p><pre>df = pd.read_csv(&quot;transactions.csv&quot;)</pre><p><strong>4. Preprocess the data:</strong></p><ul><li><em>Handle missing values</em></li><li><em>Convert categorical variables into numerical variables using one-hot encoding</em></li><li><em>Normalize the data if needed</em></li></ul><pre># Handle missing values<br>df.fillna(df.mean(), inplace=True)<br><br># Convert categorical variables into numerical variables<br>df = pd.get_dummies(df, columns=[&quot;type&quot;])<br><br># Normalize the data<br>scaler = StandardScaler()<br>df = pd.DataFrame(scaler.fit_transform(df), columns=df.columns)</pre><p><strong>5. Split the data into training and testing sets:</strong></p><pre>X = df.drop(&quot;fraud&quot;, axis=1)<br>y = df[&quot;fraud&quot;]<br>X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)</pre><p><strong>6. Train the model:</strong></p><pre>model = RandomForestClassifier()<br>model.fit(X_train, y_train)</pre><p><strong>7. Evaluate the model:</strong></p><pre>y_pred = model.predict(X_test)<br>print(classification_report(y_test, y_pred))<br>print(confusion_matrix(y_test, y_pred))</pre><p><strong>8. Make predictions on new transactions:</strong></p><pre>new_transactions = [[0.5, 0, 1, 0], [1.5, 0, 0, 1]]<br>print(model.predict(new_transactions))</pre><p><em>Please note that this is just a simple example to demonstrate the process of building a machine learning model for transaction monitoring. In a real-world scenario, the solution would be much more complex and may involve multiple algorithms, data preprocessing techniques, and feature engineering.</em></p><blockquote>There are several benefits to implementing machine learning (ML) models for transaction monitoring in order to identify anti-money laundering (AML) activities</blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LQeIWXiAkyUaC6ETUrn3Bg.png" /></figure><p><strong>1. Improved accuracy:</strong> ML models can analyze large amounts of data and detect patterns that might be missed by manual analysis, resulting in higher accuracy and fewer false negatives.</p><p><strong>2. Increased efficiency:</strong> ML models can process transactions in real-time, allowing for quicker identification of potential AML activities.</p><p><strong>3. Scalability: </strong>ML models can be easily scaled to handle increasing amounts of transaction data.</p><p><strong>4. Reducing manual workload: </strong>ML models can automate many of the tasks that are typically performed manually, freeing up time and resources for other tasks.</p><p><strong>5. Improved risk management:</strong> ML models can identify potential AML activities that might otherwise go undetected, improving risk management and reducing the risk of financial losses.</p><p><strong>6. Better compliance:</strong> ML models can help ensure compliance with AML regulations and reduce the risk of fines and penalties.</p><p><strong>7. Customizability: </strong>ML models can be easily tailored to meet the specific needs of each organization, including the ability to adapt to changing regulatory requirements.</p><h4>In conclusion, the implementation of ML models for transaction monitoring can significantly improve the accuracy and efficiency of AML activities identification, reducing the risk of financial losses and ensuring compliance with regulatory requirements.</h4><h3>NOTE:</h3><p><em>It’s important to note that while there are several publicly available datasets for various machine learning tasks, it can be challenging to find a suitable dataset for transaction monitoring and anti-money laundering (AML) activities. This is due to the sensitive nature of financial data and the need to protect sensitive information. As a result, it’s possible that a public dataset for this task might not be available.</em></p><p><em>In the absence of a public dataset, organizations can use synthetic data or anonymized data from their own transactions to train and evaluate their ML models. This data can be used to create realistic simulations that mimic the behavior of real transactions, allowing organizations to test and refine their models before deploying them in production.</em></p><p><em>It’s also important to ensure that appropriate privacy and security measures are in place when working with financial data, as well as to comply with relevant regulations, such as the General Data Protection Regulation (GDPR) in the European Union.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f1c61305cd" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Building a Machine Learning Pipeline]]></title>
            <link>https://medium.com/swlh/building-a-machine-learning-pipeline-33f512da319?source=rss-be7c33f2449b------2</link>
            <guid isPermaLink="false">https://medium.com/p/33f512da319</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[ml-pipeline]]></category>
            <category><![CDATA[machine-learnin]]></category>
            <category><![CDATA[data-science]]></category>
            <dc:creator><![CDATA[Sarath SL]]></dc:creator>
            <pubDate>Mon, 29 Jun 2020 17:49:19 GMT</pubDate>
            <atom:updated>2020-07-01T20:49:13.395Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tAiyRD1MJppOUuECm2gqRg.jpeg" /></figure><p>Artificial intelligence (AI) is wide-ranging branch of computer science concerned with building smart machines capable of performing tasks that typically require human intelligence. These tools can collect data from edge to core to cloud, analyze the data in near real time, study from previous results, and infer actions to predict future outcomes.</p><p>In current scenario, organizations implementing AI models needs to digest exponentially a massive amount of data thereby supporting complex models and simulate those models in shorter amount of time to make real time predictions.</p><p><strong><em>Below diagram depicts a simple AI pipeline diagram</em></strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/966/1*9PzRFi6pBWHcRRiitxhAZQ.png" /></figure><ul><li>We need to process many different operations that we used to prepare a Model.</li><li>Pipeline stages the transformation steps of the data and helps to make the process more standardized.</li><li>Pipeline class of sklearn helps simplifying the chaining of transformation steps and the model.</li><li>While working in a team, Pipeline regulates the model project by enforcing consistency in building a model.</li></ul><blockquote><strong><em>Build a simple ML pipeline using sklearn.pipeline</em></strong></blockquote><ol><li>First step is to import pipeline module from the sklearn.pipeline library.</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/275/1*1XwqgvYS_7uL7VsD3MJk3g.png" /><figcaption>Import</figcaption></figure><p>2. Before building the pipeline I am splitting the data into a train and test set so that I can validate the performance of the model.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/544/1*VB0qR3GLlXEEzfjshdxx1A.png" /><figcaption>Splitting Data</figcaption></figure><p>3. Now we have to define <strong>pipeline </strong>which will take transform type as function parameters.<br> I have created a numeric transformer which applies a StandardScaler and a machine learning algorithm which performs Logistic Regression.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/250/1*eXpiVIG03LX0LlxkVnJZtQ.png" /><figcaption>Defining pipeline</figcaption></figure><p>If you see above, we are providing name to each step which is being passed with the function call.<br>Alternatively we can make use of make_pipeline() function which will automatically name each step.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/494/1*DY87_bMdE-sL4vwdUnRaUg.png" /><figcaption>Using make_pipeline()</figcaption></figure><p>4. Now the next step is to call the pipeline object to fit our data.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*eBhMcF2O_5gm_rxKHAFB6g.png" /><figcaption>Fit training data</figcaption></figure><p>5. Now once the model is build, you make predictions by calling the test data.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/319/1*m7liaGmLxxSLyIIwi8M-fw.png" /><figcaption>Predict the Test data</figcaption></figure><p>Pipeline object requires all stages to have a transform() function except for the last stage which is an estimator.<br>Transforms steps transforms the input data, which will be output to the next stage of pipeline.</p><p><strong><em>Digest the above steps in a simpler way :)</em></strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/881/1*MOp965oTVhH_vYU8_4YhQA.png" /><figcaption>Reference : <a href="https://www.azuredevopslabs.com/labs/vstsextend/aml/">https://www.azuredevopslabs.com/labs/vstsextend/aml/</a></figcaption></figure><p>Thank you!!!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=33f512da319" width="1" height="1" alt=""><hr><p><a href="https://medium.com/swlh/building-a-machine-learning-pipeline-33f512da319">Building a Machine Learning Pipeline</a> was originally published in <a href="https://medium.com/swlh">The Startup</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[K Means Clustering Algorithm]]></title>
            <link>https://medium.com/@slsarath2/k-means-clustering-algorithm-5fa9d5d64326?source=rss-be7c33f2449b------2</link>
            <guid isPermaLink="false">https://medium.com/p/5fa9d5d64326</guid>
            <category><![CDATA[k-means-clustering]]></category>
            <category><![CDATA[unsupervised-learning]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Sarath SL]]></dc:creator>
            <pubDate>Wed, 30 Oct 2019 17:53:14 GMT</pubDate>
            <atom:updated>2019-10-31T03:20:44.715Z</atom:updated>
            <content:encoded><![CDATA[<p>Cluster analysis or clustering is the task of grouping a set of objects in such a way that objects in the same group are more similar to each other than to those in other groups.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/220/1*oK9astLW2l3V3Gxhq31jIg.gif" /></figure><p>Let me make it more simpler for you..<br>Whenever we visit a mall or a grocery shop to purchase items, you would have observed things are arranged in a manner that they belong to one another.<br>For ex:<br>When we look for noodles , we can see that the rack where noodles are kept also has all those items which will help to make your noodles more tastier. This make it more simpler for you to buy things related to one another. This is a simple example of clustering.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/796/1*xGcVxsLRw8rveBiu3RUCOg.png" /></figure><p>Similar kind of examples are recommendation systems used in Amazon, Netflix and other day to day digital platforms which will sort our taste according to your past history with that particular application.</p><p>There are different types of clustering techniques but in this story our main focus will be on <strong><em>“ K Means Clustering”.</em></strong></p><blockquote>K Means clustering is a type of unsupervised learning which is used to solve clustering problems. <strong>K</strong>-<strong>Means clustering</strong> intends to partition n objects into k clusters in which each object belongs to the cluster with the nearest mean.</blockquote><h4>K Means is a clustering algorithm which group similar items or data points into a cluster.<br>“K” in K means represents the number of clusters.</h4><p>We calculate the distances to assign a point to a cluster. In K-Means, each cluster is associated with a centroid. The main objective of the K-Means algorithm is to minimize the sum of distances between the points and their respective cluster centroid.</p><h3>Understanding K Means Clustering:</h3><p>We have technical support data which consists of different independent variables providing information on different types of customer related issues.<br>Using K Means clustering algorithm we will be clustering similar kind of issues into groups depending upon their problem type and other independent variables.</p><p>STEPS INVOLVED</p><ol><li>Import the necessary libraries.</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/700/1*gBA0wPn5tDjdV1tbS29TCQ.png" /></figure><p>2. Read the data and understand the data.<br>The dataset contains one record for each unique problem type. It has metrics for each type like count, average calls to resolve, average resolution time etc.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4sQssCJ-oPXdcm8YQ96uNw.png" /></figure><h3>Algorithm used in K Means clustering :</h3><p><strong>a.</strong> Clusters the data into <em>k</em> groups where <em>k</em> is predefined.<br><strong>b.</strong> Select <em>k</em> points at random as cluster centers.<br><strong>c.</strong> Assign objects to their closest cluster center according to the <em>Euclidean distance</em> function.<br><strong>d.</strong> Calculate the centroid or mean of all objects in each cluster.<br><strong>e.</strong> Repeat steps 2, 3 and 4 until the same points are assigned to each cluster in consecutive rounds.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/615/1*zhm8Ni13w7Bqgmpt7BMWHA.png" /></figure><p>3. See data distribution:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/902/1*QTSDLGgtyuKgOTkWkO9enw.png" /></figure><p>Now as mentioned in algorithm group data into similar clusters:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/914/1*rfEE2ybDCJ-s8fM9W_gTaw.png" /></figure><p>Looking into the curve , we can see possible two bends at k=3 and at k=5</p><p>Let’s take k=5 and check for the group prediction.<br>Unfortunately there is no global theoretical method to find the optimal number of clusters. A practical approach is to compare the outcomes of multiple runs with different <em>k</em> and choose the best one based on a predefined criterion. In general, a large <em>k</em> probably decreases the error but increases the risk of overfitting.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/946/1*lC4Ivx_URjSpMTOOnWcoGg.png" /></figure><h3>Analyze the distribution of the data when K = 5</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/880/1*DBDtEg8ABIr6bmQt4UGaCw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/844/1*xJTCSn0xb_8uwTXatY5J6Q.png" /></figure><p>So by using K Means and k=5, we have identified 4 different set of groups/cluster.</p><p>The initial choice of centroids can affect the output clusters, so the algorithm is often run multiple times with different starting conditions in order to get a fair view of what the clusters should be.</p><p>Check out Python code <a href="https://github.com/slsarath/Machine-Learning/tree/master/KMeansClustering"><strong>HERE</strong></a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5fa9d5d64326" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[PAIRPLOT VISUALIZATION]]></title>
            <link>https://medium.com/analytics-vidhya/pairplot-visualization-16325cd725e6?source=rss-be7c33f2449b------2</link>
            <guid isPermaLink="false">https://medium.com/p/16325cd725e6</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[seaborn]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[exploratory-data-analysis]]></category>
            <dc:creator><![CDATA[Sarath SL]]></dc:creator>
            <pubDate>Sun, 29 Sep 2019 10:27:13 GMT</pubDate>
            <atom:updated>2020-07-02T17:32:41.215Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="PAIRPLOT VISUALIZATION" src="https://cdn-images-1.medium.com/max/1024/1*aSC3odScNMyz7Y6MZvqJ1Q.jpeg" /></figure><p>Pairplot visualization comes handy when you want to go for Exploratory data analysis (“EDA”).</p><p>Pairplot visualizes given data to find the relationship between them where the variables can be continuous or categorical.<br><strong><em> Plot pairwise relationships in a data-set.</em></strong></p><p>Pairplot is a module of <strong><em>seaborn</em> </strong>library which provides a high-level interface for drawing attractive and informative statistical graphics.</p><p>Let’s see how the same has been done.</p><ol><li>Get your data</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/434/1*XktSK0L8gle8O5VLHKLHwA.png" /></figure><p>As observed from the data , you can see there are both continuous and categorical columns in data-set.</p><p>Let’s plot data using pairplot:<br>From the picture below, we can observe the variations in each plot. The plots are in matrix format where the row name represents x axis and column name represents the y axis. The main-diagonal subplots are the univariate histograms (distributions) for each attribute.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/784/1*7SvQlVBBbbp_FLTQD5iPFQ.png" /></figure><p><strong>Pairplot Parameters:<br>seaborn.pairplot(<em>data</em>, <em>hue=None</em>, <em>hue_order=None</em>, <em>palette=None</em>, <em>vars=None</em>, <em>x_vars=None</em>, <em>y_vars=None</em>, <em>kind=&#39;scatter&#39;</em>, <em>diag_kind=&#39;auto&#39;</em>, <em>markers=None</em>, <em>height=2.5</em>, <em>aspect=1</em>, <em>dropna=True</em>, <em>plot_kws=None</em>, <em>diag_kws=None</em>, <em>grid_kws=None</em>, <em>size=None</em>)</strong></p><p>Except data, all other parameters are optional. There are few other parameters which <strong>pairplot</strong> can accept.</p><p>Getting understanding on some commonly used parameters — →</p><h3>different levels of a categorical variable by the color of plot elements:</h3><p>Hue helps you to get the difference in variable in data to map plot aspects to different colors.<br><em>sns.pairplot(df, hue=”smoker”)</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/731/1*4o0cVRK2iCFgyE5aGxv4EA.png" /></figure><p>Use a different color palette.<br>Its basically a set of colors for mapping the hue variable.<br><strong><em>g = sns.pairplot(df, hue=”smoker”, palette=”husl”)</em></strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/642/1*geYwovAI291-69DVomyYfg.png" /></figure><p>Use different markers for each level of the hue variable:<br><strong><em>sns.pairplot(df, hue=”smoker”, markers=[“o”, “s”])</em></strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/656/1*6n96PQ9YCeBza7OwTsuA0Q.png" /></figure><p>We can also plot subplots using pairplot using “vars”.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/312/1*Ap8ffCY7sRdpAllJKINi1g.png" /></figure><p>Commonly used:<br><strong><em>sns.pairplot(df,hue = ‘smoker’,diag_kind = “kde”,kind = “scatter”,palette = “husl”)</em></strong></p><p>where :<br><strong>kind : </strong>Kind of plot for the non-identity relationships. {‘scatter’, ‘reg’}<br><strong>diag_kind : </strong>Kind of plot for the diagonal subplots. {‘hist’, ‘kde’}</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/695/1*jY9TFv7k9rEpaPqGqydoog.png" /></figure><p>Pair plot is used to understand the best set of features to explain a relationship between two variables or to form the most separated clusters. It also helps to form some simple classification models by drawing some simple lines or make linear separation in our data-set.</p><p>There are other different plots by which you can do EDA technique which helps you to get better understanding on the relationship among different variables of the data.</p><p>All of the above techniques will help you to take best suited variables to build your Machine Learning Model.</p><p>Thank you!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=16325cd725e6" width="1" height="1" alt=""><hr><p><a href="https://medium.com/analytics-vidhya/pairplot-visualization-16325cd725e6">PAIRPLOT VISUALIZATION</a> was originally published in <a href="https://medium.com/analytics-vidhya">Analytics Vidhya</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>