<p>Quantifying the suspended sediment load (SSL) in Iran’s Gheshlagh basin, specifically the Chehel Gazi (1) and Khalifa Tarkhan (2) Rivers, is crucial for ensuring Sanandaj’s water security and regional ecological stability. This research pioneers an innovative two-stage approach, which employs data classification (DC) based on machine learning (ML) to enhance the predictive accuracy and robustness of traditional sediment rating curve (SRC) and advanced artificial neural network (ANN) methods in sediment modeling. Phase 1 (Classification): Water-sediment (W-S) discharge data were classified using Gaussian mixture models (GMM) and Naive Bayes classifiers (NBC). Specifically, 19 GMMs (2–20 classes) were evaluated using the Bayesian information criterion (BIC) and Akaike information criterion (AIC) to determine the optimal number of classes. This facilitated a comparative analysis between unclassified data (Mode 1) and classified data (Mode 2). Phase 2 (Modeling): classified and unclassified W-S data were used to develop SRC and ANN models for SSL estimation (70% training, 30% testing). In Mode 2, NBC assigned test data to GMM-identified classes. Results indicated that GMM identified two optimal classes for both rivers. Notably, AI-driven DC significantly improved the predictive accuracy of both the SRC and ANN models. ANN consistently outperformed SRC in both modes, with Mode2-ANN achieving the highest accuracy. Compared to traditional Mode1-SRC, Mode2-ANN reduced errors by 90.3% for Riv. 1 and 37.3% for Riv. 2. AI-driven DC was more effective at reducing errors than the modeling methods (ANN, SRC), and its integration with regression techniques proves vital for improved hydrological predictions, especially in data-scarce regions where traditional methods falter.</p>