Table 3 Comparative Analysis from the review perspective.
From: Reliable water quality prediction and parametric analysis using explainable AI models
Reference | Algorithms | Input parameters | Evaluation results |
|---|---|---|---|
PNN | BOD, PO4-P, COD, temperature, NO 3-N, Ca 2+, Cl-, alkalinity, P, Mg 2+, pH,and EC | Interpolation is good performance - R2 : 0.82 | |
BWNN, ANN, ARIMA, BANN | Dissolved Oxygen | ARIMA < ANN<WNN <BANN<BWNN | |
LSTM | Dissolved Oxygen | High runoff ratio \(\ge\) 0.45 \(\bullet\) - 74% of sites | |
CCNN | water quality and DO parameters (example: Cl, NO x,pH, TDS, and temperature) | R2 : 0.825 RMSE: 0.550 | |
SVM, ANN | TDS, Na+, Mg 2+,Temperature, pH, EC, HCO3 , Cl, and Ca 2+ | SVM performs better than ANN | |
SVM, ANN | flow travel time, rainfall, river flow, temperature, DO, TN, and TP | SVM perform better than ANN | |
DT, RF, DCF, and 10 other models | pH, DO, CONMn, and NH 3-N | DCF, DT, and RF performed well | |
SVR | EC, fDOM, turbidity, BGA-PC, chlorophyll-a, DO, and sediments | BGA-PC : (Accuracy: 0.77), chlorophyll-a (Accuracy: 0.78), TSS (Accuracy: 0.81), from (-), turbidity (Accuracy: 0.55) | |
Attention-based neural network | Images of water | Accuracy: Polluted-water= 73.6% Accuracy: Clean-water=71.2% | |
SVM, RF, CNN | Landsat8 images | RF Accuracy: 86.21% SVM Accuracy: 96.89% CNN Accuracy: 97.12% |
