Optimizing ETo Prediction in the Mahanadi Basin: A Comprehensive Evaluation of Machine Learning Models with Emphasis on ANFIS Performance

Abstract

Abstract This study extensively analyzed three models, M5P, ANFIS, and GEP, to predict Actual Evapotranspiration (ETo) in the Mahanadi Basin region on six major stations Raipur, Korba, Jharsuguda, Bilaspur, Bhubaneswar, and Balangir. Evaluation metrics, including R2, RMSE, NSE, and MAE, were applied to a testing dataset, revealing ANFIS's consistent superiority with high R2 (0.930746 to 0.990526) and NSE (0.926792 to 0.990458) values, alongside the lowest RMSE (0.101152 to 0.332819) and MAE (0.000386 to 0.034319). Weighted scores affirmed ANFIS's dominance across multiple stations, except for specific instances where GEP excelled in Bhubaneswar and M5P in Balangir. The study highlighted ANFIS's proficiency in predicting ETo values at specific locations, demonstrated through effective variation capture in scatter plots. The discussion underscored the importance of model selection, emphasizing the versatility of machine learning models and the effectiveness of combining AI techniques for accurate ETo prediction. ANFIS consistently outperformed M5P and GEP, solidifying its status as a reliable ETo prediction tool. While acknowledging M5P and GEP's potential in specific contexts, the study stressed the need to tailor models to unique location characteristics. Reference to related studies supported the effectiveness of hybridized AI approaches in improving ETo modeling. The study advocated ongoing research to refine models, incorporate additional factors, and enhance predictive accuracy. The findings contribute valuable insights for water resource management, irrigation planning, and agricultural decision-making across diverse locations.