Preparation of water quality dataset using outlier detection and imputation methods

Reliable water quality assessment depends on high-quality datasets; however, datasets obtained from field measurements often contain outliers and missing values, which directly affect subsequent analytical processes. This study presents a practical system application that enhances dataset integrity and reliability for advanced artificial intelligence analyses. The proposed approach integrates exploratory visualization (boxplots) for outlier detection, statistical outlier treatment, mean- and regression-based estimation for missing values, and normalization for ensuring comparability. The application utilizes a river water quality dataset from Türkiye, including Cl⁻, Fe, K⁺, Na⁺, SO₄²⁻, TKN, TN, and turbidity parameters. Model performance demonstrated strong results, with Mean Squared Error (MSE) values for the identified outliers ranging from 0.00002 (turbidity) to 0.135 (K⁺). Comparative analysis of raw and post-processed datasets revealed that systematic outlier handling and targeted imputation improved data consistency and reduced modeling uncertainty, thereby enabling more reliable ANN- and GA-based predictive modeling. The proposed methodological framework is practical, reproducible, and easily integrable into water quality monitoring systems, supporting data-driven management and policy decisionmaking.