Predicting the inflow of the Euphrates River using machine learning approaches

Water resource management is of paramount importance for nations, playing a critical role in sustaining ecosystems and supporting diverse economic sectors. However, it comes with challenges such as escalating demand, climate change, and water scarcity. Artificial intelligence (AI) has emerged as a promising tool to address these challenges and enhance water resource management practices. Yet, predicting and understanding water resource behaviour for national decision-making requires a specialized methodology beyond the study of natural phenomena. This research introduces a comprehensive methodology that surpasses traditional approaches and applies to water resource problems. We applied the methodology to aid the decision-making in the Syrian Ministry of Water Resources by predict the Euphrates River’s inflow, entails collaboration with experts from data collection initiation to model application. It ensures accurate feature selection, preventing the model from being purely theoretical and making it practical. The model considers a wide range of factors, including economic and political conditions represented by oil prices, alongside natural factors like temperature, precipitation, and seasonality. The methodology uses three classification algorithms (OneR, Naïve Bayes, and Random Tree) with four data splitting techniques (80/20 split, 66/34 split, K-Fold Cross-Validation and Random Splitting with 10 Iterations approach) during training and testing phases for a more comprehensive evaluation. The results shows that Naïve Bayes algorithm achieved the highest classification accuracy in the K-fold approach, reaching 84.52%. On the other hand, the convergence in the results ranging from 79.77% to 86.12% underscores the methodology’s practicality, aligning predictions with expert knowledge and emphasizing its reliability for effective water resource management.