Analyzing non-revenue water dynamics in Rwanda: leveraging machine learning predictive modeling for comprehensive insights and mitigation strategies

Abstract

This study investigated non-revenue water (NRW) dynamics in Rwanda from 1 July 2014 to 30 June 2023, utilizing panel data and cross-sectional datasets. It aimed to assess progress toward achieving the government's 25% NRW reduction target. Through panel data analysis and machine learning models, it examined water supply variations, NRW levels, and associated risks across fiscal years and regions. The observed average NRW of 41.24% underscores the need for targeted interventions to meet the set target. Regional disparities, exemplified by Kigali City's water network's 38.61% average NRW compared to Nyagatare's 55.31%, emphasize the importance of tailored strategies. Machine learning models indicated low and inconsistent progress across networks. Notably, no single water supply managed to meet the target in more than 20% of the 36 quarters studied. Comparison with existing literature highlighted excessive NRW in Rwanda, aligning with global trends. Achieving the 25% NRW target requires region-specific approaches, necessitating infrastructure improvements, leak detection, and capacity building. The positive correlation between water loss risk and household access to improved water sources accentuated the complexity in NRW reduction efforts. This study contributes to understanding NRW dynamics and informs sustainable water management strategies tailored to Rwanda's context.