Validating a KPI/MDT-based methodology for RAN optimization

Addressing the gap that no KPI directly measures indoor traffic percentage, this thesis validates MDT-based methodology developed at TIM company to estimate indoor outdoor traffic percentage at cell and hour level and assesses its usefulness for RAN optimization. Two full day datasets from February and May 2025 are analyzed by performing data characterization and developing statistical learning model. To inspect data, correlations with selected KPIs are computed, cell clustering depends on location is performed, user mapping and RSRP analyses are conducted and Generalized Additive Model is fitted with smooth KPI functions, hour of day term and SINRxPRB interaction. Correlation results show that radio quality metrics increase with indoor share while load and throughput decrease. Although the aggregate correlation values are lower then expected, clustering reveals much stronger relationships. Spatial analyses align with physical expectations, indoor dominant city center cell has lower average RSRP (likely due to penetration losses), whereas highway cells are outdoor dominant with RSRP driven bu traffic proximity. The GAM reproduces the diurnal pattern with good in-sample accuracy (MAE 0.033–0.050), with performance improving for clusters with more cells. Tests on unseen cell tests has higher error values (MAE ≈0.206–0.210), but preserve the overall shape reasonably well. These findings support the indicator as an actionable input for coverage and capacity tuning, and motivate hierarchical models and richer covariates to better capture cell level heterogeneity.