Using Machine Learning To Predict Rainfall in Abeokuta Nigeria

Abstract

Rainfall is extremely important in Abeokuta as it supports agriculture, influences daily life, and the local economy. Abeokuta receives substantial seasonal precipitation; however, variability poses significant risks. This work focuses on Abeokuta and examines rainfall variability from 2001 to 2010. Using machine learning models is one way to extract information, patterns, and trends from historical data, enabling stakeholders to make informed decisions. It is essential to measure the climatic variables that correlate with rainfall, as this will help identify ways to mitigate their effects in pursuit of a balanced nature. In this study, the Seasonal Autoregressive Integrated Moving Average (SARIMA), Random Forest (RF), and Artificial Neural Network (ANN) were used to examine seasonality and patterns in our dataset for predictive purposes. The results show that the ANN model's MSE (7767.4691) and MAE (64.7500) are the lowest among the models used in this work. This suggests that the ANN model's predictions are closer to the actual rainfall values than those of other models. Furthermore, a correlation analysis revealed that the ANN correlated with evaporation and relative humidity. This suggests that during wet seasons or prolonged rainy periods, when rainfall is high, evaporation is low, and less irrigation may be needed, but soils stay wet longer. This might result in flooding. In contrast, in dry seasons with little rain, evaporation is high, and soils dry out quickly, increasing irrigation demand.