Forecasting Water Demand, Distribution Efficiency and Consumption Patterns in Sokoto Metropolis Using a Hybrid ARIMA-ANN Model

Abstract

Water scarcity in rapidly urbanizing semi-arid regions necessitates precise forecasting for sustainable management. This study develops and applies a novel hybrid Autoregressive Integrated Moving Average-Artificial Neural Network (ARIMA-ANN) model to forecast water demand, distribution efficiency, and consumption patterns in Sokoto Metropolis, Nigeria. Utilizing monthly data (2018–2024) from the Sokoto State Ministry of Water Resources, the research compares the performance of standalone ARIMA, standalone ANN, and the proposed hybrid model. Evaluation metrics Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE) consistently demonstrated the hybrid model's superiority. The hybrid model achieved a MAPE of 2.57% for water demand, 1.65% for distribution efficiency, and 1.57% for consumption rate, outperforming both individual models. Forecasts for 2025–2026 indicate a steady rise in water demand from 335.23 to 404.93 million litres, while distribution efficiency (78.40%) and consumption rate (83.48%) remain stagnant, highlighting critical inefficiencies. The study concludes that the hybrid ARIMA-ANN framework provides a robust, accurate tool for urban water resource planning, offering actionable insights for infrastructure investment and policy formulation in water-stressed cities.