Complementing missing stream data with artificial intelligence algorithms managed by heuristic optimization methods

In this thesis study, Artificial Neural Network (ANN) models governed by heuristic optimization algorithms were developed to complete missing streamflow data. Using data from eight streamflow gauging stations (1401, 1412, 1413, 1414, 1424, 1428, 1409, and the target station 1402) located in the Yeşilırmak Basin of Türkiye, missing data were estimated. ANN was used as the basic model, and three different heuristic optimization algorithms—Particle Swarm Optimization (PSO), Grey Wolf Optimization (GWO), and Ant Colony Optimization (ACO)—were integrated to improve the learning and generalization performance of the model. Each hybrid model was optimized through 30 repeated runs with different particle and iteration numbers (5–10, 10–50, 50–100). The performance of the models was evaluated using MAE, RMSE, and R² metrics, and Taylor diagrams and flow hydrographs were used to visually demonstrate the prediction stability. The results showed that, compared to the classical ANN model, the hybrid structures generally produced lower error values and higher correlation coefficients. In particular, the PSO–ANN and GWO–ANN models stood out in terms of the balance between accuracy and stability, while the ACO–ANN model provided the highest generalization performance in the testing stage with the highest R² and the lowest RMSE values. Station-based analyses revealed that stations 1413, 1424, and 1428 made the highest contributions to model accuracy. In addition, it was observed that models using five or more input variables significantly reduced error levels. The findings showed that optimization algorithms are effective in determining model hyperparameters and that different heuristic structures significantly improve model performance. This study reveals the effectiveness of ANN operated by heuristic optimization in completing missing streamflow data in the field of hydrological modeling.