Optimizing Fixed Broadband Market Penetration Strategies with XGBoost and Geographic Visualization

Fixed broadband penetration in Indonesia has grown slowly in the past two decades, with only a 1.2 % increase from 2001 to 2011. By 2020, penetration had reached 15%, yet the market remains largely unexplored, presenting substantial opportunities for fixed-broadband providers. Identifying and predicting high-potential market areas is essential for designing effective market penetration strategies. This study identifies and analyzes factors that influence the high take-up ratio (TUR) by performing feature engineering and evaluating predictive models of machine learning (ML). Several baselines were analyzed, including a heuristic approach, Logistic Regression, XGBoost, Multilayer Perceptron (MLP), Decision Tree, and Support Vec-tor Machine (SVM). Model performance was evaluated using accuracy, precision, recall, Fl-Score, and ROV AUC. The best performing models underwent hyperparameter tuning and were integrated with ArcGIs for spatial visualization. The results showed that XGBoost outperformed other models after hyper-parameter optimization, with an ROC AUC score of 0.76 and an accuracy of 0.705. The predictions of the optimized model were visualized using ArcGIS, which showed the Tangerang areas with a high TUR. This study identifies XGBoost as an effective model for predicting high TUR areas. This integrated approach offers a data-driven framework for market penetration decisions based on high TUR area predictions combined with geographic visualization. Future research should improve model accuracy by exploring feature engineering from additional data sources, such as customer after sales datasets.