COMPARATIVE ANALYSIS ON HYBRID MACHINE LEARNING ALGORITHMS: AN APPROACH TO RAINFALL PREDICTION IN MAIDUGURI AND PORTHARCOURT

This study presents a comparative analysis of hybrid machine learning algorithms for rainfall prediction in two climatically diverse Nigerian cities: Maiduguri (a semi-arid region) and Port Harcourt (a humid tropical region). The research developed and evaluated four hybrid models (SWLR-BAT, SWLR-BRT, SWLR-GPR, and SWLR- SVM) by combining stepwise linear regression (SWLR) with non-linear algorithms, including BAT, BRT, GPR, and SVM. Monthly rainfall data spanning 33 years (1990- 2022) were obtained from the NASA power dav, incorporating seven meteorological variables. The methodology employed a systematic approach where linear baseline predictions were combined with non-linear residual modeling to capture complex meteorological patterns. Two variable groups were evaluated for each station: group 1, comprising the four most statistically correlated variables with precipitation, and group 2 containing four physically relevant meteorological parameters. Performance evaluation utilized five statistical metrics: BIAS, MAD, MSE, RMSE, and. Results demonstrated exceptional improvements in hybrid models over standalone algorithms, with SWLR-GPR achieving the highest validation performance in both stations, at 0.9487 for Maiduguri and 0.8987 for port harcourt. This study conclusively validates that integrating a hybrid approach effectively models complex meteorological dynamics and provides a robust framework for improving forecasting tools essential for Nigeria’s agricultural sector, disaster preparedness, and water resource management.