Comparative Forecasting of Rainfall in Nigeria Using ARIMA and Artificial Neural Networks

This study presents a comparative analysis of two prominent time series forecasting models, Autoregressive Integrated Moving Average (ARIMA) and Artificial Neural Networks (ANN), in predicting monthly rainfall in Nigeria. Utilizing a quantitative methodology, the research analyzes average monthly rainfall data from January 1980 to April 2025, obtained from the Humanitarian Data Exchange. The dataset, known for its linear and nonlinear patterns, was preprocessed through linear interpolation and normalization to ensure compatibility with each modeling technique. The ARIMA model was tuned using AIC and BIC to identify optimal lag structures after differencing to attain stationarity. ANN employed a feedforward architecture with a single hidden layer trained via backpropagation. Both models were evaluated using Root Mean Square Error (RMSE) across varying training sample sizes (40%, 60%, 80%, and 100%) to test generalizability. Results indicate that while ARIMA offers interpretability and model parsimony through AIC and BIC, it consistently reported higher RMSE values (7.5–8.1) compared to ANN (0.276–0.284). This demonstrates ANN’s superior forecasting accuracy, particularly with larger datasets. Despite ARIMA’s strength in explainability, ANN proves more efficient in capturing nonlinear rainfall patterns and minimizing forecast errors. The study underscores ANN as a more effective tool for rainfall prediction 
in data environments typical of developing nations.