Table 1 State-of-the-art ML models employed for wastewater parameters.

Support vector machines

Real wastewater treatment plant (AMBEV WWTP)

SVM

COD

29.94

30.86

− 11.97

Auto optimization of ML hyperparameters

²¹

VFCW

SVR

NH₄-N

1.35

0.89

XAI approaches

¹⁰

TN

3.64

0.90

TP

0.08

0.87

Real-time South African municipal WWTP

SVR

Conductivity (Con)

8.35

50.47

–

XAI approaches for interpretation

²⁰

COD

21.04

24.86

–

TSS

12.16

16.22

–

Neural networks (ANN, LSTM, BiLSTM, DNN)

Real-time South African municipal WWTP

LSTM

Conductivity (Con)

2.28

50.21

–

XAI approaches for interpretation

²⁰

BiLSTM

COD

1.54

24.89

–

Umeå WWTP

DNN

Phosphate in effluent (PO4e)

–

–

0.872

Incorporating ML interpretation frameworks

²³

Lab-scale two-staged A/O WWTP

LSTM

Influent NH₄-N

–

0.101

–

Auto optimization of ML hyperparameters

²⁴

The spectral reflectance dataset

ANN (multiple targets)

Influent BOD

2.415

–

0.973

Auto optimization of NN hyperparameters

¹⁴

Tree-based models (RF, CatBoost)

Real wastewater treatment plant (AMBEV WWTP)

RF

COD

30.03

30.95

− 3.30

Auto optimization of ML hyperparameters

²¹

Moscoe Region, Russia WWTP

CatBoostRegressor

BOD (Effluent)

0.670

0.962

0.78

²²

COD (Effluent)

0.854

1.211

0.85

Ammonium (Effluent)

0.111

0.145

0.79

Phosphorus (Effluent)

0.036

0.054

0.81

Nitrate (Effluent)

0.484

0.765

0.82

Nitrite (Effluent)

0.003

0.006

0.88

Umeå WWTP

RF

PO4e

–

–

0.886

Incorporating ML model interpretation systems to enhance the framework and improve the reliability of its results.

²³

TSSe

–

–

0.920

Regression models (MLR, rPCA-MARS)

Lab-scale two-staged A/O WWTP

MLR

Influent NH₄-N

–

0.203

–

²⁴

The spectral reflectance dataset

Linear rPCA-MARS

BOD

9.879 (MSE)

–

0.959

XAI approaches for interpretation

²⁵

COD

213.072 (MSE)

–

0.967

NH₃-N

3.032 (MSE)

–

0.936

Cubic rPCA-MARS

BOD

12.508 (MSE)

–

0.939

COD

284.445 (MSE)

–

0.945

NH₃-N

3.113 (MSE)

–

0.934