Title: Comparative analysis of machine learning algorithms efficacy in water resources management

Authors: Danijela Voza; Goran Babić; Milovan Vuković; Ivana Mladenović Ranisavljević

Addresses: Technical Faculty in Bor, University of Belgrade, Vojske Jugoslavije 12, 19210 Bor, Serbia ' Faculty of Business and Law, MB University, Teodora Drajzera 27, 11040 Belgrade, Serbia ' Technical Faculty in Bor, University of Belgrade, Vojske Jugoslavije 12, 19210 Bor, Serbia ' Faculty of Technology in Leskovac, University of Niš, Bulevar Oslobodjenja 124, Leskovac, Serbia

Abstract: One of the most reliable indicators of surface water quality is dissolved oxygen (DO). This study aims to propose a single model and the optimal number of variables for reliable DO prediction. In this regard, a comparative analysis of the efficiency of representative machine learning models in DO prediction was conducted. The initial parameters for examining the possibility of predicting the DO concentration were pH, temperature (T), electrical conductivity (EC), ammonium (NH₄) and orthophosphates (PO₄). Four input combinations of these water quality parameters were created based on the variable importance analysis, and machine learning techniques were applied to each subset. The obtained values indicated that all the predictive models perform best when four water quality variables - T, pH, PO₄, and EC are used as input. Also, according to the results, the best-fitted model on the created dataset is support vector machine, and the deep learning model slightly lags behind it.

Keywords: dissolved oxygen; prediction; machine learning algorithms; water quality.

DOI: 10.1504/IJEWM.2025.149522

International Journal of Environment and Waste Management, 2025 Vol.38 No.3, pp.295 - 323

Received: 28 Sep 2023
Accepted: 30 Jan 2024
Published online: 05 Nov 2025 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article