Kolmogorov-Arnold Networks for Interpretable Analysis of Water Quality Time Series Data

Kolmogorov–Arnold Networks (KAN) represent a promising model suited for applications that require interpretability. Here, we explore the use of KAN to analyze time series of water quality parameters obtained from a published dataset related to an aquaponic environment. The Water Quality Index (WQI) was calculated as an arithmetic combination of three water parameters: pH, total dissolved solids, and temperature. By training KAN models, we derived explicit algebraic expressions capable of accurately predicting WQI, achieving low prediction error while emphasizing the most relevant predictors. Model performance was assessed using standard regression metrics, with R2 values exceeding 0.97 on the test set. These findings highlight the potential of KAN and its applicability to broader problems where accuracy, interpretability, and model simplification are desirable.