Development of a Hybrid Machine Learning Model for Predicting Determinant Variables in Wastewater Contamination and Assessing Its Agricultural Reuse Potential in Mexico

Abstract

Water scarcity has become one of the most critical challenges for sustainable agricultural development, particularly in semi-arid regions where water availability limits agricultural productivity. In this context, treated wastewater reuse has emerged as a viable alternative to supplement conventional water resources. However, fluctuations in wastewater quality complicate the timely assessment of its suitability for agricultural applications. This study proposes a novel hybrid machine learning framework, termed the Hybrid Intelligent Agricultural Suitability Model for Wastewater Reuse (HIASM-WR), designed to predict key contamination variables and automatically classify the agricultural reuse potential of treated wastewater. The proposed architecture integrates Random Forest, Extreme Gradient Boosting (XGBoost), Long Short-Term Memory (LSTM) neural networks, fuzzy logic, and Explainable Artificial Intelligence (XAI) techniques through SHAP and LIME. Based on an extensive review of recent scientific literature and the specific challenges of water management in Mexico, the model was designed to simultaneously predict physicochemical, microbiological, and environmental parameters while generating an Agricultural Suitability Index (ASI). The results indicate that the proposed framework addresses limitations identified in previous studies by incorporating Mexican environmental regulations and providing an interpretable decision-support tool for wastewater reuse. The study contributes a novel methodological approach to strengthen sustainable water management strategies in agriculture and support climate adaptation efforts in water-stressed regions.