Modeling of diatom indices (Bdı, Tdı and Gdı) based on the physico-chemical structure of the river ecosystem with machine learning and artificial intelligence methods; a comparative example

Abstract Diatom indices are used to assess the quality of aquatic plants in sustainable river ecosystems. The traditional assessment of diatom indices involves complicated and lengthy process steps. Today, artificial intelligence-based modelling plays a key role in overcoming this complexity. The aim of this work is to model selected diatom indices Biological Diatom Index (BDI), Trophic Diatom Index (TDI) and General Diatom Index (GDI) based on the physicochemical structure of river ecosystems using artificial intelligence and machine learning methods. The application part of the study used surface water variables from rivers monitored by 5 different stations for 24 months as a data set. Traditional analyses were compared with artificial intelligence and machine learning methods using the MATLAB programme. Different algorithms were considered, including Neural Network/Multilayer Perceptron (MLP), Support Vector Machine (SVM), Linear Regression (LR), Gaussian Process Regression (GPR), Decision Tree and Levenberg-Marquardt (LM) approach. To evaluate the quality of the models, the coefficient of determination (R 2 ), root mean square error squared (RMSE) and mean absolute percentage error (MAPE) were compared. The R 2 values of the Levenberg-Marquardt model, which gave the best prediction results for BDI, TDI and GDI, were found to be Validation; 0.7691, Training; 0.9620 Testing; 0.8613, Validation 0.9273, Training; 0.9303, Testing; 0.9199, Validation; 0.9273, Training; 0.9303, Testing; 0.9199, respectively. Levenberg Marquardt efficiently predicted Diatom index results accurately with high precision. Our results show that artificial intelligence and machine learning methods are highly efficient tools for the prediction of diatom indices. A time-efficient and labour-saving application in sustainable ecosystem management was successfully demonstrated.