Predicting FTTH Take-Up Ratio in West Java Using SHAP and LIME Methods

Internet Service Providers (ISPs) invest in broadband networks, covering network architecture, infrastructure, services, and user terminals, mainly funded by CAPEX. Evaluating these investments involves assessing the take-up rate, average user rate, and post-development profitability. Current methodologies for measuring take-up rates have two main drawbacks: reliance on small time-series data and lack of correlation between components and expected results. This study aims to identify predictive algorithms that explain input elements influencing outcomes and build models with minimal training data. Using CRM FTTH data from West Java (2019-2020), five regression models were evaluated, with SVR outperforming linear regression, ANN, RBFN, and ELM in RMSE and MAE metrics. Results were interpreted using SHAP for SVR and LIME for LR. Both methods highlighted the importance of competition and price in predicting take-up rates, with SHAP providing comprehensive insights due to its global granularity and cumulative impact. Despite LIME indicating the significance of users over 55 and business locations, SHAP’s detailed interpretation showed competition’s substantial impact. Further research should use national datasets and improve RBFN and ELM models to handle limited data volumes effectively.