River pollution caused by municipal and domestic sewage is an escalating global issue, exacerbated by rapid urbanisation and population growth, particularly in developing countries like Sri Lanka. This study developed a basin-specific sewage pollution index for the Kelani River Basin, using advanced machine learning (ML) techniques. The index was developed in four stages: selection of parameters, sub-indexing, sub-index weighting, and sub-index aggregation. Five parameters were selected to develop the index based on measurements taken from 2016 to 2020. Water quality standards established by the Central Environmental Authority of Sri Lanka were considered to develop rating curves for each of the five parameters to obtain sub-index values. The random forest model (AUC = 0.9357, F1-score = 0.9313) and the rank-order centroid method assigned the parameter weights in the same order of importance. A sensitivity analysis was conducted for five weighted sub-index aggregation functions to determine the best-fitting method. The selected method was validated through an LSTM-ANN model (NSE = 0.993, RMSE = 1.681). The obtained pollution values were categorised into five classes using the percentile method. Among 17 monitoring stations, the findings indicated that Maha Ela, Raggahawatte Ela, Seethawake Ferry, Kollonnawa Bridge, Victoria Bridge, and Pugoda Ela are consistently extremely polluted throughout the year. The study's findings highlight the critical need for countermeasures and policy interventions to address the sewage pollution in the Kelani River Basin, providing a scalable methodology for other river basins facing similar challenges.