<?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 Mochamad Satria Tatas Dirangga on Medium]]></title>
        <description><![CDATA[Stories by Mochamad Satria Tatas Dirangga on Medium]]></description>
        <link>https://medium.com/@ranggadangga?source=rss-ccb3b75df66b------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*KL5MVn3P3KFF36QI-55_Dw.jpeg</url>
            <title>Stories by Mochamad Satria Tatas Dirangga on Medium</title>
            <link>https://medium.com/@ranggadangga?source=rss-ccb3b75df66b------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 21:43:01 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@ranggadangga/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[Which Credit Card Customers Are Most at Risk? Insights from K-Means Clustering]]></title>
            <link>https://medium.com/@ranggadangga/which-credit-card-customers-are-most-at-risk-insights-from-k-means-clustering-7c9f708b52f2?source=rss-ccb3b75df66b------2</link>
            <guid isPermaLink="false">https://medium.com/p/7c9f708b52f2</guid>
            <category><![CDATA[college-students]]></category>
            <category><![CDATA[watches]]></category>
            <category><![CDATA[habits]]></category>
            <category><![CDATA[dataset]]></category>
            <category><![CDATA[calculations]]></category>
            <dc:creator><![CDATA[Mochamad Satria Tatas Dirangga]]></dc:creator>
            <pubDate>Thu, 19 Nov 2026 07:25:14 GMT</pubDate>
            <atom:updated>2026-11-19T07:25:14.305Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*QAibmPk1ncf9uShdjKdDlA.jpeg" /><figcaption><strong>K-Means Clustering summary for credit card customer risk segmentation.</strong> <em>(Top Left)</em> Elbow (inertia) and Silhouette score evaluation across $k = 2$ to $10$, supporting $k = 4$ as a practical choice for policy design. <em>(Top Right)</em> PCA projection ($k = 4$) showing how the four clusters sit along a behavioral spectrum with gradual overlap rather than rigid boundaries. <em>(Bottom)</em> Standardized (<em>z-score</em>) feature heatmap across clusters, contrasting disciplined payers (Cluster 0) against cash advance users (Cluster 1), light spenders (Cluster 2), and burdened revolvers (Cluster 3).</figcaption></figure><p>Many banks treat credit card customers as one single homogeneous group. In reality, some customers pay their balances in full every month, others use their cards solely for cash advances, and some have balances nearly brushing against their credit limits. Giving everyone the same offers and collection treatments can cause banks to miss the mark — either being too lenient with risky customers or too intrusive with good ones.</p><p>In this article, I map customers using K-Means Clustering, but from a credit risk management perspective rather than purely marketing. My question is simple: Which customer groups exhibit the most concerning behavior, and what can banks do about it?</p><h3>Data and Limitations</h3><p>I used the <em>Credit Card Dataset for Clustering</em> from Kaggle (uploaded by arjunbhasin2013). The dataset summarizes the usage behavior of roughly 9,000 active credit card holders over 6 months, with 18 behavioral variables per customer — specifically, 8,950 customers. The dataset does not name the bank and has no default labels, so “risk” here refers to behavioral indicators, not proof that customers have actually defaulted. The data is also very likely not from an Indonesian bank, so the findings are better read as behavioral patterns rather than a snapshot of the Indonesian market. github</p><p>There were 313 missing values in the minimum payments column and 1 in credit limit. I imputed them with the median. Upon inspection, 77 percent of those missing rows had zero total payments and a median balance of only around 17, meaning most were nearly inactive accounts.</p><h3>Engineered Features</h3><p>I did not feed all 17 raw columns directly into the model. Many columns overlap and are heavily right-skewed. I created six features more closely tied to risk behavior:</p><ul><li>Credit utilization, calculated as balance divided by credit limit (capped at a maximum of 1.5).</li><li>Ratio of payments to minimum payments (log-transformed).</li><li>Share of cash advance relative to total cash advance plus purchases.</li><li>Cash advance frequency.</li><li>Purchase frequency.</li><li>Percentage of full payment.</li></ul><p>All features were then standardized (z-score) before entering K-Means.</p><h3>Choosing the Number of Clusters</h3><p>I tested $k$ from 2 to 10 using the elbow method and silhouette score (Rousseeuw, 1987). The highest silhouette score was indeed at $k = 2$ (0.334), but two clusters are too coarse for policy purposes. I chose $k = 4$. The elbow bend sits around that number, the silhouette score is 0.301, the Davies-Bouldin index is 1.18, and all four clusters can be clearly interpreted.</p><p>A silhouette score of 0.30 is moderate. This means the boundaries between clusters are not as sharp as textbook diagrams, resembling a behavioral spectrum instead. To test stability, I re-ran K-Means on 30 random subsamples containing 80 percent of the data. Agreement with the main result was very high (average Adjusted Rand Index of 0.99 and minimum of 0.98), so these clusters are not a one-off fluke.</p><p>(Insert fig1_elbow_silhouette.png here.)</p><h3>Four Customer Profiles</h3><p>(Insert fig2_profil_klaster.png here.)</p><p><strong>Cluster 0, Disciplined Payers</strong> (1,345 customers, 15.0 percent). Their average utilization is only 4.7 percent, purchase frequency is high (0.79), and 78.5 percent of bills are paid in full. They hold just 2.0 percent of total balances, yet account for 29.7 percent of total purchases. Almost none pay below the minimum (1.4 percent). These are the healthiest customers.</p><p><strong>Cluster 1, Cash Advance Users</strong> (3,100 customers, 34.6 percent). Their cash advance share is 93.5 percent and they barely make purchases (purchase frequency of 0.135). Only 3.2 percent pay in full, and 41.7 percent pay below the minimum. This group holds 54.9 percent of total balances and 84.9 percent of total cash advances, but only 7.5 percent of total purchases. The bank’s largest balance burden sits here.</p><p><strong>Cluster 2, Light Spenders</strong> (2,939 customers, 32.8 percent). Their utilization is low (12.8 percent), purchase frequency is fairly high (0.61), and they have almost no cash advances. They contribute 38.0 percent of total purchases. However, only 7.2 percent pay in full and 16.1 percent pay below the minimum, so they need monitoring to prevent drifting into Cluster 3.</p><p><strong>Cluster 3, Burdened Revolvers</strong> (1,566 customers, 17.5 percent). Their utilization is the highest (79.2 percent), median credit limit is the lowest (2,100 compared to 3,000 to 4,000 in other clusters), and almost nobody pays in full (0.7 percent). As many as 54.3 percent pay below the minimum. They make up only 17.5 percent of customers, yet hold 29.7 percent of total balances. This is the most vulnerable group.</p><p>(Insert fig3_pca.png here. The first two principal components explain 67.6 percent of the data&#39;s variance.)</p><h3>What This Means for Banks</h3><p>Below are sensible policy recommendations based on the patterns above. These are analytical proposals, not field-tested results.</p><ul><li><strong>Cluster 3:</strong> Priority for early monitoring. Examples include earlier due-date reminders, installment restructuring offers before delinquency occurs, and careful credit limit reviews.</li><li><strong>Cluster 1:</strong> Monitor reliance on cash advances given their very large share of balances. Educating customers on cash advance fees and offering lower-cost loan products could serve as alternatives.</li><li><strong>Cluster 2:</strong> Encourage larger and automated payments. This group drives spending, so the goal is keeping them active without getting trapped in revolving balances.</li><li><strong>Cluster 0:</strong> Retain with loyalty programs. Intensive collections are unnecessary.</li></ul><h3>A Little Indonesian Context</h3><p>In Indonesia, overall banking credit quality remains relatively sound. OJK recorded a banking gross NPL ratio of 2.07 percent and net NPL of 0.82 percent in September 2026. However, consumer financing based on digital installments is growing rapidly. Banking paylater grew 33.54 percent year-on-year to Rp30.71 trillion in June 2026, with an NPL of 2.36 percent. These figures are aggregate and not specific to credit cards, so they cannot be directly compared with the clusters above. Still, paylater growth highlights the importance of early behavioral segmentation before non-performing loan ratios rise. bisnisinfobanknews</p><h3>Limitations</h3><ol><li>There are no default labels, so risk indicators are merely behavioral proxies.</li><li>The “paying below minimum” indicator uses the median for 313 missing minimum payment records (around 3.5 percent of the data).</li><li>K-Means assumes spherical clusters and is sensitive to outliers. A silhouette score of 0.30 shows that boundaries between clusters are not sharp.</li><li>The bank and country of origin of the data are unknown, limiting generalization to Indonesia.</li></ol><p>An interesting next step is testing how these clusters relate to actual default data and comparing the results against other methods such as Gaussian Mixture Models.</p><h3>References</h3><ul><li>Kaggle. <em>Credit Card Dataset for Clustering</em> (arjunbhasin2013). <a href="https://www.kaggle.com/datasets/arjunbhasin2013/ccdata">https://www.kaggle.com/datasets/arjunbhasin2013/ccdata</a></li><li>MacQueen, J. (1967). Some methods for classification and analysis of multivariate observations. <em>Proceedings of the Fifth Berkeley Symposium on Mathematical Statistics and Probability</em>, 1, 281–297.</li><li>Pedregosa, F., et al. (2011). Scikit-learn: Machine learning in Python. <em>Journal of Machine Learning Research</em>, 12, 2825–2830.</li><li>Rousseeuw, P. J. (1987). Silhouettes: a graphical aid to the interpretation and validation of cluster analysis. <em>Journal of Computational and Applied Mathematics</em>, 20, 53–65.</li><li>Bisnis.com (2026). <em>OJK: Kredit Perbankan Tumbuh 13,26%, Kualitas Aset Tetap Terjaga</em>. <a href="https://finansial.bisnis.com/read/20261103/90/2018070/ojk-kredit-perbankan-tumbuh-1326-kualitas-aset-tetap-terjaga">https://finansial.bisnis.com/read/20261103/90/2018070/ojk-kredit-perbankan-tumbuh-1326-kualitas-aset-tetap-terjaga</a></li><li>Infobanknews (2026). <em>Warga RI Makin Doyan Pakai Paylater, Ini Buktinya</em>. <a href="https://infobanknews.com/warga-ri-makin-doyan-pakai-paylater-ini-buktinya">https://infobanknews.com/warga-ri-makin-doyan-pakai-paylater-ini-buktinya</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7c9f708b52f2" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Comparing ARIMA, RNN, LSTM, and GRU for Stock Price Forecasting
Time series analysis of Bank…]]></title>
            <link>https://medium.com/@ranggadangga/comparing-arima-rnn-lstm-and-gru-for-stock-price-forecasting-time-series-analysis-of-bank-04218e8394d2?source=rss-ccb3b75df66b------2</link>
            <guid isPermaLink="false">https://medium.com/p/04218e8394d2</guid>
            <category><![CDATA[saham]]></category>
            <category><![CDATA[investing]]></category>
            <category><![CDATA[timeseries]]></category>
            <category><![CDATA[data-collection]]></category>
            <category><![CDATA[data-science]]></category>
            <dc:creator><![CDATA[Mochamad Satria Tatas Dirangga]]></dc:creator>
            <pubDate>Wed, 28 Oct 2026 07:41:46 GMT</pubDate>
            <atom:updated>2026-10-28T07:41:46.521Z</atom:updated>
            <content:encoded><![CDATA[<h3>Comparing ARIMA, RNN, LSTM, and GRU for Stock Price Forecasting<br>Time series analysis of Bank Mandiri and Apple stock prices using Python 3</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/736/1*htMat8DfbfshxVyqWwYMjA.jpeg" /><figcaption>this article inspired by this book</figcaption></figure><h3>Introduction</h3><p>Stock price movements are a prime example of time series data, which consists of data points collected sequentially over time. Stock prices fluctuate due to numerous factors, such as economic conditions, corporate performance, market sentiment, and government policies. Consequently, forecasting stock prices presents an interesting problem for analysis using both statistical and machine learning methods.</p><p>In this study, four models are compared to predict daily closing stock prices: ARIMA (Autoregressive Integrated Moving Average), Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU). The analysis is conducted using two stocks: Bank Mandiri (BMRI.JK), representing the Indonesian stock market, and Apple Inc. (AAPL), representing the international stock market. After evaluating the models, the model with the lowest Mean Absolute Percentage Error (MAPE) is selected to forecast stock prices over the next two trading weeks.</p><h3>Data and Data Sources</h3><p>The dataset consists of daily closing prices (Close) over the last five years, calculated automatically relative to the execution date of the notebook. Data is downloaded from Yahoo Finance using the yfinance Python library. The stock tickers used are BMRI.JK for Bank Mandiri and AAPL for Apple Inc.</p><p>The download date and timestamp are recorded directly by the program. Therefore, the reported download date reflects the actual execution output of the notebook. Stock price data is only available on market trading days. Consequently, weekends and holidays contain no price observations. These missing dates are not filled with artificial values.</p><h3>Tools and Environment</h3><p>The analysis is conducted using Python 3 within the Google Colab environment. The key libraries include:</p><ul><li>yfinance for retrieving stock market data.</li><li>pandas and NumPy for data manipulation and processing.</li><li>Matplotlib for data visualization.</li><li>statsmodels for Augmented Dickey-Fuller (ADF) testing and ARIMA modeling.</li><li>scikit-learn for feature scaling and evaluation metrics.</li><li>TensorFlow/Keras for building the RNN, LSTM, and GRU models.</li></ul><h3>1. Data Exploration</h3><p>The first step involves downloading and inspecting the closing price data for each stock. Missing entries are removed, and dates are sorted chronologically. Daily closing price plots are then generated to observe general trends, price variations, and periods of high volatility.</p><p>Descriptive statistics are also calculated, including the number of observations, mean, standard deviation, minimum value, quartiles, and maximum value. These statistics provide an initial overview of the price distribution. However, they are insufficient on their own to explain the underlying causes of stock price movements.</p><h3>2. Stationarity Test with ADF</h3><p>Stationarity is tested using the Augmented Dickey-Fuller (ADF) test. The null hypothesis (H0) states that the series contains a unit root, meaning that it is non-stationary. If the p-value is less than 0.05, the null hypothesis is rejected, and the series is considered stationary at the 5 percent significance level.</p><p>The ADF test is performed on both the original closing price series (Close) and its first-order differenced series. First-order differencing is calculated using the following formula:</p><p>Delta P(t) = P(t) — P(t-1)</p><p>Where:</p><p>P(t) = Closing price at time t.</p><p>P(t-1) = Closing price on the previous observation day.</p><p>Delta P(t) = Change in the closing price between two consecutive observations.</p><p>The ADF results should be interpreted directly from the notebook output table. If the original price series is non-stationary but the first-differenced series is stationary, it indicates that differencing is required for the ARIMA model. The final conclusion must be based on the actual p-value obtained from the analysis.</p><h3>3. Train-Test Data Splitting</h3><p>The dataset is divided chronologically into two subsets. The first 80 percent of observations are used as the training set, while the remaining 20 percent are used as the test set.</p><p>The data is not shuffled because temporal order is an essential characteristic of time series analysis. Identical data splits are applied to all models for each stock to ensure a fair comparison.</p><p>The training set is used to develop the models, while the test set evaluates their ability to predict previously unseen observations. For neural network models, MinMaxScaler is fitted exclusively to the training data to reduce the risk of data leakage from the test set.</p><h3>4. Models Compared</h3><h3>4.1 ARIMA (Autoregressive Integrated Moving Average)</h3><p>ARIMA is a statistical model that combines three components: autoregressive (p), differencing (d), and moving average (q). The model is represented as ARIMA(p,d,q).</p><p>In the notebook, several simple parameter combinations are compared based on their Akaike Information Criterion (AIC) values calculated using the training set. The model with the lowest AIC is selected from the tested combinations.</p><h3>4.2 RNN (Recurrent Neural Network)</h3><p>RNN is a neural network architecture designed to process sequential data. It uses information from previous observations to help predict subsequent values. However, traditional RNNs may struggle to retain long-term dependencies when processing long sequences.</p><h3>4.3 LSTM (Long Short-Term Memory)</h3><p>LSTM is an advanced type of recurrent neural network that uses gating mechanisms to retain, update, and filter information over time. These mechanisms help the model learn long-term dependencies, making LSTM suitable for many time series forecasting problems.</p><h3>4.4 GRU (Gated Recurrent Unit)</h3><p>GRU is another type of recurrent neural network that uses a simplified gating structure compared with LSTM. This design can reduce computational requirements and allow more efficient training in certain situations.</p><p>For the RNN, LSTM, and GRU models used in this notebook, the previous 20 observation days are used as input to predict the closing price for the next observation day. All models are trained using the same training dataset and evaluated over the same test period.</p><h3>5. Model Evaluation: RMSE and MAPE</h3><p>Model performance is evaluated using Root Mean Squared Error (RMSE) and Mean Absolute Percentage Error (MAPE).</p><p>RMSE measures the magnitude of prediction errors in the same units as the stock prices. MAPE measures the average absolute prediction error as a percentage of the actual values.</p><p>RMSE Formula:</p><p>RMSE = sqrt[(1/n) * Sum((y_i — y_hat_i)²)]</p><p>MAPE Formula:</p><p>MAPE = (100/n) * Sum(abs((y_i — y_hat_i)/y_i))</p><p>Where:</p><p>y_i = Actual stock price.</p><p>y_hat_i = Predicted stock price.</p><p>n = Total number of evaluated observations.</p><p>sqrt = Square root.</p><p>Sum = Summation of all evaluated observations.</p><p>Lower RMSE and MAPE values generally indicate better predictive accuracy. In this study, MAPE is used as the primary criterion for selecting the best-performing model for each stock.</p><p>The evaluation tables should be inserted directly from the notebook execution output. Numerical results must not be filled in before the models have been executed. Once the results are available, the RMSE and MAPE values of ARIMA, RNN, LSTM, and GRU should be compared for both BMRI.JK and AAPL.</p><p>Actual-versus-predicted plots should also be included to visually assess how closely each model’s predictions follow the actual stock price movements.</p><h3>6. Two-Week-Ahead Forecasting</h3><p>After model evaluation, the model with the lowest MAPE is selected separately for each stock. The selected model is then retrained using the full five-year dataset so that it can utilize all available historical observations before generating forecasts.</p><p>The forecast covers the next 10 business days, representing approximately two trading weeks.</p><p>ARIMA generates multi-step predictions directly from the fitted model. In contrast, RNN, LSTM, and GRU use recursive forecasting. This means that each predicted value is fed back into the input sequence to generate the next prediction.</p><p>The forecast results are presented in a table containing the estimated dates and predicted closing prices. A corresponding plot displays the last 60 historical observations alongside the 10 forecast points.</p><p>Forecast dates from Monday through Friday represent estimated business days. Market holidays are not explicitly accounted for and should be verified before publication.</p><h3>7. Results and Discussion</h3><p>This section should be completed after executing the notebook.</p><p>For BMRI.JK, the model with the lowest MAPE should be identified. The RMSE and MAPE values for all four models should be recorded and compared. The discussion should also explain whether the predicted values closely follow the actual stock price movements.</p><p>The same evaluation should be conducted for AAPL. The results should identify which model performs best for each stock based on the test set.</p><p>Next, the 10-business-day forecast table for each stock should be inserted. The predicted price trajectory can then be described as an upward, downward, or sideways trend, depending on the actual forecast results.</p><p>Forecasts should not be presented as guaranteed future outcomes because the models provide estimates based on historical patterns rather than certainty about future market conditions.</p><p>Example of a results statement after the numerical values are available:</p><p>“For stock [ticker], the model achieving the lowest MAPE on the test set was [Model Name], with an RMSE of [value] and a MAPE of [value] percent. Based on the MAPE criterion, this model was selected for forecasting. The 10-business-day forecast indicates a [trend description] trend, as observed in the forecast chart.”</p><h3>Conclusion</h3><p>This study compares the statistical ARIMA model with three neural network architectures, namely RNN, LSTM, and GRU, to forecast the closing prices of Bank Mandiri and Apple stocks.</p><p>The analysis includes data exploration, descriptive statistics, ADF stationarity testing, chronological 80:20 train-test splitting, model evaluation using RMSE and MAPE, and a 10-business-day forecast using the best-performing model.</p><p>The study does not assume that one model will consistently outperform the others across all stocks and forecasting periods. Instead, the optimal model is determined through empirical evaluation using the test dataset. A more complex model does not necessarily produce more accurate predictions.</p><p>Furthermore, stock price forecasting has inherent limitations because the models rely exclusively on historical closing prices. Other potentially important factors, such as trading volume, financial reports, news sentiment, government policies, and macroeconomic indicators, are not included in the analysis.</p><p>The findings of this study are intended strictly for academic purposes and do not constitute investment advice. Actual stock prices may differ significantly from forecasted values because of unpredictable market dynamics and other factors that cannot be captured by the models.</p><h3>References and Reproducibility Notes</h3><ol><li><a href="https://www.google.com/finance/">Google Finance — Harga Pasar Saham, Kuotasi Real-time &amp; Berita Bisnis</a></li><li><a href="https://www.investing.com/">Investing.com — Stock Market Quotes &amp; Financial News</a></li><li><a href="https://finance.yahoo.com/">Yahoo Finance — Stock Market Live, Quotes, Business &amp; Finance News</a><br>Description</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/951/1*Tolcl642mkToXQV-sTNfMw.png" /><figcaption>Data Information and Sources</figcaption></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=04218e8394d2" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Stop Overcomplicating Databases: How I Built a Live Regional Government Dashboard Using Google…]]></title>
            <link>https://medium.com/@ranggadangga/stop-overcomplicating-databases-how-i-built-a-live-regional-government-dashboard-using-google-f3017c85f5bf?source=rss-ccb3b75df66b------2</link>
            <guid isPermaLink="false">https://medium.com/p/f3017c85f5bf</guid>
            <category><![CDATA[dashboard]]></category>
            <category><![CDATA[transactions]]></category>
            <category><![CDATA[distributed-transaction]]></category>
            <category><![CDATA[web-development]]></category>
            <category><![CDATA[netlify]]></category>
            <dc:creator><![CDATA[Mochamad Satria Tatas Dirangga]]></dc:creator>
            <pubDate>Sun, 18 Oct 2026 06:53:57 GMT</pubDate>
            <atom:updated>2026-10-18T06:53:57.888Z</atom:updated>
            <content:encoded><![CDATA[<h3>Stop Overcomplicating Databases: How I Built a Live Regional Government Dashboard Using Google Sheets, Netlify,</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/735/1*Em01vdu9wE7LjEE5YkJd8Q.jpeg" /><figcaption>Fetching spreadsheet data and dynamically rendering it into HTML DOM elements requires robust JavaScript logic</figcaption></figure><h3>Introduction: Learning by Building Real Things</h3><p>Most beginner-to-intermediate developers get stuck before writing a single line of code. Why? Because we tend to overcomplicate the stack — worrying about setting up complex databases, backend servers, and authentication before even testing our idea.</p><p>I decided to take a different path: <strong>Learn by doing, ship fast, and keep it simple.</strong></p><p>Recently, I completed a hands-on project: <strong>Building an interactive ETPD (Regional Transaction Electronification Readiness) Dashboard for 13 Local Governments in East Java</strong> (including Kediri, Blitar, Madiun, Ponorogo, Pacitan, Ngawi, Trenggalek, and more).</p><p>The goal was to transform qualitative indicators and quantitative estimates from a spreadsheet into a clean, interactive, and publicly accessible web dashboard.</p><p>You can check out the live dashboard here:</p><p>👉 <a href="https://dashboardconnectkesiapanetpdpemda13.netlify.app/"><strong>Live ETPD Readiness 13 Pemda Dashboard</strong></a></p><p>In this post, I’ll walk you through the entire journey — from setting up the architecture, step-by-step implementation, the frustrating failures I hit along the way, and how <strong>Claude AI</strong> acted as my coding co-pilot to pull it all off.</p><h3>The Lightweight Tech Stack</h3><p>Rather than building a heavy backend, I opted for a lean, efficient stack:</p><ol><li><strong>Google Sheets</strong> — Acts as the flexible, zero-cost database storing the 12 Commitments × 13 Regional Governments matrix.</li><li><strong>HTML5 / CSS3 / JavaScript</strong> — Provides a lightweight, responsive, and interactive frontend.</li><li><strong>Netlify</strong> — Offers instant, friction-free continuous deployment for static web hosting.</li><li><strong>Claude AI</strong> — Served as my pair programmer for writing integration logic, debugging errors, and cleaning raw API responses.</li></ol><h3>Step-by-Step Implementation</h3><h4>Step 1: Structuring Data in Google Sheets</h4><p>I organized the data into structured sheets containing qualitative readiness markers (Yes/No flags) and quantitative index baselines for the 13 local governments. To make the data readable by my web app, I published the sheet to the web and exposed it as a fetchable endpoint.</p><h4>Step 2: Designing the Interactive Frontend</h4><p>I crafted a clean layout featuring executive summaries, dynamic filters for each local government, indicator breakdowns, and visual status cards.</p><h4>Step 3: Writing Integration Logic with Claude AI</h4><p>Fetching spreadsheet data and dynamically rendering it into HTML DOM elements requires robust JavaScript logic (data parsing, array transformations). I teamed up with <strong>Claude AI</strong> to draft the fetch logic and map the spreadsheet rows into structured JSON objects.</p><h4>Step 4: One-Click Deployment on Netlify</h4><p>Once the frontend code and data fetching scripts were tested locally, I pushed the codebase to Netlify. Within seconds, Netlify generated a live public URL complete with free SSL encryption.</p><h3>The Ugly Part: Failures, Bugs, and Trial-and-Error</h3><p>No project goes smoothly on the first try. Here are the main roadblocks I ran into and how I solved them:</p><h4>1. The Dreaded CORS &amp; Access Denied Error</h4><ul><li><strong>The Failure:</strong> When I opened my local browser console, the dashboard was completely blank, spewing CORS (<em>Cross-Origin Resource Sharing</em>) and error messages.</li><li><strong>The Root Cause:</strong> The Google Sheet wasn’t configured with public reading permissions, and the initial endpoint URL format was blocked by browser security policies.</li><li><strong>The Fix:</strong> With Claude AI’s guidance, I adjusted the sharing permissions to <em>“Anyone with the link can view”</em> and updated the fetch URL to use a clean public CSV/JSON format that bypasses CORS restrictions.</li></ul><h4>2. Messy Data Parsing &amp; Header Offset Bugs</h4><ul><li><strong>The Failure:</strong> The fetched data was rendering out of order — values meant for “Kota Kediri” were showing up under “Kab. Blitar”, and header notes were displaying as data rows.</li><li><strong>The Root Cause:</strong> The original Google Sheet included descriptive note rows above the actual table headers, causing row-index mismatches.</li><li><strong>The Fix:</strong> Claude AI helped me write a data-sanitization helper function in JavaScript that detected non-data rows, applied index offsets, and dynamically mapped columns by header names instead of hardcoded array positions.</li></ul><h4>3. Broken State Re-renders on Filtering</h4><ul><li><strong>The Failure:</strong> Whenever a user selected a different Pemda from the dropdown menu, the new data appended below the old data instead of updating the view.</li><li><strong>The Fix:</strong> Claude AI pointed out that I forgot to clear the DOM container before re-rendering, and supplied a simple state-reset snippet to keep the interface snappy.</li></ul><h3>How Claude AI Accelerated the Development Process</h3><p>Working with <strong>Claude AI</strong> transformed how I approached problem-solving throughout this project:</p><ul><li><strong>Instant Error Diagnosis:</strong> Instead of spending hours searching through stack trace forums, I pasted raw console error messages into Claude AI, and it immediately pinned down the exact line causing the issue.</li><li><strong>Clean Async Data Fetching:</strong> Claude AI generated modern, easy-to-read JavaScript functions wrapped in robust blocks to gracefully handle network failures or slow connections.</li><li><strong>UI Component Generation:</strong> Claude AI assisted in styling responsive CSS cards and badges so that status indicators (“Yes”, “No”, “Estimates”) adapted cleanly across mobile and desktop devices.</li></ul><h3>Key Takeaways</h3><ol><li><strong>You Don’t Need a Complex Stack to Ship Value:</strong> Google Sheets combined with modern JavaScript is more than enough to power interactive client dashboards quickly.</li><li><strong>AI Belongs in Your Workflow:</strong> Claude AI didn’t write the project <em>for</em> me — it acted as an expert mentor, explaining <em>why</em> bugs happened and teaching me cleaner code practices along the way.</li><li><strong>Embrace Failures:</strong> Debugging CORS issues and data parsing errors taught me far more about how the web works than a smooth tutorial ever could.</li></ol><p>Be sure to explore the live dashboard here:</p><p>👉 <a href="https://dashboardconnectkesiapanetpdpemda13.netlify.app/"><strong>https://dashboardconnectkesiapanetpdpemda13.netlify.app/</strong></a></p><p><em>How are you utilizing AI tools like Claude AI in your development workflow? Have you ever used Google Sheets as a database? Let’s connect and chat in the comments below!</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f3017c85f5bf" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Boundless Haze: Re-evaluating Ecological Crisis and Peatland Fires in Kalimantan]]></title>
            <link>https://medium.com/@ranggadangga/boundless-haze-re-evaluating-ecological-crisis-and-peatland-fires-in-kalimantan-d403e4991b43?source=rss-ccb3b75df66b------2</link>
            <guid isPermaLink="false">https://medium.com/p/d403e4991b43</guid>
            <category><![CDATA[environment]]></category>
            <category><![CDATA[public-health]]></category>
            <category><![CDATA[environmental-issues]]></category>
            <category><![CDATA[goverment]]></category>
            <category><![CDATA[climate-impacts]]></category>
            <dc:creator><![CDATA[Mochamad Satria Tatas Dirangga]]></dc:creator>
            <pubDate>Wed, 02 Sep 2026 10:13:03 GMT</pubDate>
            <atom:updated>2026-09-02T10:13:03.801Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/675/1*FL6eUwWK4SkWBk3P-VxQSw.jpeg" /><figcaption>“Beyond the flames: Tracing the ecological roots and governance challenges behind seasonal forest and peatland fires.”</figcaption></figure><p>With every dry season, Kalimantan’s landscape faces a recurring crisis: forest and peatland fires (<em>karhutla</em>). The dense haze enveloping local communities is far more than a routine natural occurrence or an inevitable consequence of extreme weather. Beneath the suffocating smoke lies a complex web of land conversion, deforestation, and systemic governance gaps.</p><p>Analyzing the wildfires in Kalimantan requires a broader lens — moving beyond the immediate flames on the ground to trace the ecological, social, and policy roots that perpetuate this environmental catastrophe.</p><h3>Chronology of a Crisis: From Wetlands to Wastelands</h3><p>Canalization &amp; Peatland Drainage</p><p>Phase 1</p><p>Large-scale land clearing alters the natural hydrology of peat ecosystems. Constructing drainage canals lowers the water table drastically, leaving underground peat layers dry, degraded, and hyper-flammable.</p><p>Seasonal Triggers &amp; Uncontrolled Burnings.</p><p>Phase 2</p><p>During dry spells or El Niño cycles, low-cost land clearing using fire triggers widespread hotspots. Flames spread rapidly beneath the surface (<em>subsurface fires</em>), making containment extremely difficult.</p><p>Transboundary Haze &amp; Public Health Emergencies.</p><p>Phase 3</p><p>Thick toxic smoke halts regional economic activity, grounds transportation, and causes sharp spikes in acute respiratory infections ($URTI$), often crossing international borders into neighboring countries.</p><h3>Multidimensional Impacts</h3><ul><li><strong>Permanent Peatland Degradation:</strong> Subsurface peat fires burn slowly underground and are notoriously difficult to extinguish. Full ecosystem restoration requires decades, if not centuries.</li><li><strong>Health &amp; Demographic Strain:</strong> Long-term exposure to fine particulate matter ($PM_{2.5}$) severely compromises public health, disproportionately affecting children and the elderly while eroding daily labor productivity.</li><li><strong>Regional Economic Losses:</strong> Agriculture, logistics, aviation, and hospitality suffer substantial financial hits due to airspace closures and halted supply chains.</li><li><strong>Biodiversity Loss:</strong> Destruction of critical habitats pushes endemic species, including the Bornean orangutan, closer to extinction and forces them into human settlements.</li></ul><h3>Stakeholder Architecture &amp; Action Plan</h3><pre>┌─────────────────────────────────────────┐<br>                   │    Wildfire Governance &amp; Restoration    │<br>                   └────────────────────┬────────────────────┘<br>                                        │<br>        ┌───────────────────────────────┼───────────────────────────────┐<br>        ▼                               ▼                               ▼<br>┌───────────────┐               ┌───────────────┐               ┌───────────────┐<br>│ State &amp; Legal │               │ Private Sector│               │  Communities  │<br>│  Regulators   │               │ (Concessions) │               │   &amp; Local NGO │<br>└───────┬───────┘               └───────┬───────┘               └───────┬───────┘<br>        │                               │                               │<br>┌───────┴───────┐               ┌───────┴───────┐               ┌───────┴───────┐<br>│- Strict legal │               │- Peat rewet-  │               │- Participatory│<br>│  enforcement  │               │  ting (REW)   │               │  monitoring   │<br>│- Moratorium on│               │- Strict zero- │               │- Sustainable  │<br>│  new permits  │               │  deforestation│               │  traditions   │<br>└───────────────┘               └───────────────┘               └───────────────┘</pre><ul><li><strong>Rigorous Enforcement &amp; Corporate Accountability:</strong> Imposing transparent civil and criminal penalties on concession owners responsible for illegal burning or land neglect.</li><li><strong>Hydrological Peat Restoration:</strong> Prioritizing canal blocking to rewet degraded peat domes, ensuring moisture levels remain high year-round.</li><li><strong>Community-Led Early Detection:</strong> Integrating local community patrols with satellite-based early warning systems to neutralize hotspots before they escalate.</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d403e4991b43" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Unraveling the Resilience of Indonesia’s Banking Giant: A Financial & Efficiency Analysis of PT…]]></title>
            <link>https://medium.com/@ranggadangga/unraveling-the-resilience-of-indonesias-banking-giant-a-financial-efficiency-analysis-of-pt-5ea8e6607e01?source=rss-ccb3b75df66b------2</link>
            <guid isPermaLink="false">https://medium.com/p/5ea8e6607e01</guid>
            <category><![CDATA[analytics]]></category>
            <category><![CDATA[finance-and-banking]]></category>
            <category><![CDATA[data-analysis]]></category>
            <dc:creator><![CDATA[Mochamad Satria Tatas Dirangga]]></dc:creator>
            <pubDate>Tue, 01 Sep 2026 12:29:07 GMT</pubDate>
            <atom:updated>2026-09-01T12:29:07.518Z</atom:updated>
            <content:encoded><![CDATA[<h3>Unraveling the Resilience of Indonesia’s Banking Giant: A Financial &amp; Efficiency Analysis of PT Bank Central Asia Tbk (BBCA)</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yX0HcfRjMNTIcgi4nmeCaA.jpeg" /></figure><p><strong>“Why Does BBCA Remain the Gold Standard of Indonesian Banking?”</strong></p><p>Amid global interest rate fluctuations and macroeconomic shifts, Indonesia’s banking sector faces a continuous challenge: maintaining operational efficiency while sustaining long-term growth. If there is one institution that consistently serves as the national benchmark for financial resilience, PT Bank Central Asia Tbk (BBCA) almost always leads the conversation.</p><p>BBCA’s dominance is not merely a product of its massive asset size. Instead, it stems from an unmatched ability to command low-cost funding (<em>CASA</em>), maintain pristine credit quality, and drive technology-led operational efficiency. This article breaks down BBCA’s core financial indicators to examine how its transaction banking ecosystem serves as its strongest fortress in the national financial market.</p><h3>1. The Core Foundation: CASA Dominance and Transaction Banking</h3><ul><li><strong>Focus:</strong> Analyzing the proportion of <em>Current Account Savings Account</em> (CASA) relative to total Third-Party Funds (DPK).</li><li><strong>Key Takeaway:</strong> How processing high daily transaction volumes keeps cost of funds exceptionally low, protecting profitability even during aggressive interest rate cycles.</li></ul><h3>2. Efficiency &amp; Financial Performance Metrics</h3><p><strong>Financial RatioPurpose &amp; Analytical InterpretationNIM</strong> (<em>Net Interest Margin</em>)Measures net interest earned from lending relative to total interest-bearing assets.<strong>CIR / BOPO</strong> (<em>Cost-to-Income Ratio</em>)Evaluates operational efficiency (a lower ratio indicates higher efficiency).<strong>ROA / ROE</strong>Assesses management’s effectiveness in generating returns from total assets and equity.<strong>LDR</strong> (<em>Loan-to-Deposit Ratio</em>)Indicates liquidity management and prudence in converting deposits into loans.</p><h3>3. Asset Quality &amp; Risk Management</h3><ul><li><strong>Focus:</strong> Examining <em>Non-Performing Loan</em> (NPL) ratios and NPL coverage buffers.</li><li><strong>Key Takeaway:</strong> How conservative underwriting standards enable sustainable credit expansion without compromising balance sheet strength.</li></ul><h3>4. The Digital Moat: Ecosystem Transformation</h3><ul><li><strong>Focus:</strong> Analyzing the shift from branch-based transactions to digital channels (myBCA, BCA mobile).</li><li><strong>Key Takeaway:</strong> How technology expenditure operates not merely as an expense, but as a strategic moat that locks in user engagement and scales operational efficiency.</li></ul><h3>5. Key Takeaways</h3><ul><li>Concluding synthesis detailing strategic lessons from BBCA’s business model: Funding Efficiency, Asset Prudence, and Scalable Digital Infrastructure.</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5ea8e6607e01" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Growing Up in the Shadows of Deception: How I Learned to Trust Again After My Parents Betrayed Me]]></title>
            <link>https://medium.com/@ranggadangga/growing-up-in-the-shadows-of-deception-how-i-learned-to-trust-again-after-my-parents-betrayed-me-c7ea9361ae09?source=rss-ccb3b75df66b------2</link>
            <guid isPermaLink="false">https://medium.com/p/c7ea9361ae09</guid>
            <category><![CDATA[childhood-trauma]]></category>
            <category><![CDATA[personal-narrative]]></category>
            <dc:creator><![CDATA[Mochamad Satria Tatas Dirangga]]></dc:creator>
            <pubDate>Tue, 01 Sep 2026 05:32:48 GMT</pubDate>
            <atom:updated>2026-09-01T05:32:48.066Z</atom:updated>
            <content:encoded><![CDATA[<p>Hi, my name is Mochamad Satria Tatas Dirangga — you can call me Satria.</p><p>This is a piece of my life story. It is an honest confession of what it feels like to grow up without ever truly having a home, and how childhood trauma carved the person I am today.</p><h3>The Illusory Beginning</h3><p>It all started with my parents. They met as neighbors in the same residential complex, fell in love, and eventually tied the knot. From the outside, everything seemed picture-perfect. My father appeared to be a deeply pious and respectable man, while my mother was a devoted, hard-working woman.</p><p>However, those masks quickly slipped away.</p><p>My first wound was inflicted before I even understood what pain was. When I was just a toddler barely a year old, my mother risked her life giving birth to me all by herself. Meanwhile, my father was right outside the hospital, busy gambling on card games. He only stepped in for a brief moment to name me, and then walked away.</p><h3>A Home Turned Battlefield</h3><p>From that moment on, our house dissolved into chaos. Financial and moral storms battered our lives relentlessly. My father’s gambling addiction escalated into infidelity, and he went so far as to sell our household furniture to fuel his vices. Left with no choice, my mother fought alone to keep food on the table, selling groceries in the local wet market.</p><p>My childhood memories are not of bedtime stories, but of screaming matches, glass shattering against the floor, and objects flying across the room. Back then, I would shrink into a corner, keeping quiet and head bowed. I genuinely believed this violence was normal — just a regular routine every family went through.</p><h3>The Final Collapse</h3><p>The breaking point arrived when I was in the fourth grade. My father was arrested by the police for drug trafficking. But while he was behind bars, another bitter betrayal shattered what remained of our family: my mother began an affair with my father’s close friend — repeating the very same betrayal my father had committed.</p><p>Watching everything fall apart stripped away my faith in humanity. I began to question: <em>Is love between two people even real, or is it just a strategic game of mutual exploitation?</em></p><p>From that day forward, I slammed the door on romantic affection. To me, love was nothing more than hypocrisy wrapped in sweet, empty promises.</p><h3>The Lone Wolf Defense</h3><p>As the years went by, I built a thick wall around myself, adopting the persona of a <strong>“lone wolf.”</strong> I convinced myself that I didn’t need anyone. I believed I could handle everything on my own — no shoulder to cry on, no vent sessions, and no asking for help.</p><p>Yet, it took me a long time to realize that this hyper-independence was just a self-imposed illusion. I was lying to myself to keep from getting hurt again. Deep down beneath the armor, I was still a human being longing for safety, warmth, and genuine love.</p><h3>To Anyone Carrying a Similar Burden</h3><p>If you are reading this while carrying childhood wounds of your own, I want to share a few lessons I learned along the way:</p><ol><li><strong>Do not be afraid to lean on others.</strong> Find at least one person who makes you feel safe and genuinely cared for. Depending on someone isn’t a sin, and asking for help doesn’t make you weak. We are human, and we were built to connect.</li><li><strong>Acknowledge your own resilience.</strong> You have been incredibly strong to survive and make it this far. Whenever you feel exhausted or feel like giving up on life, look back at the relentless storms you’ve already weathered just to stand where you are today.</li></ol><p>You are not alone, and it is completely okay not to be okay all the time.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c7ea9361ae09" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>