Evaluate effect of 126 pre-processing methods on various artificial intelligence models accuracy versus normal mode...

Area of Science:

• Hydrology and Water Resources Engineering
• Artificial Intelligence in Environmental Science

Background:

• Accurate estimation of groundwater levels (GWLs) is critical for sustainable water resource management.
• The Hamedan-Bahar plain in west Iran faces challenges in water resource management, necessitating improved GWL prediction.

Purpose of the Study:

• To compare the effectiveness of various data pre-processing techniques on artificial intelligence (AI) models for predicting groundwater levels (GWLs).
• To evaluate machine learning (ML), deep learning (DL), and hybrid-ML models for their predictive accuracy in GWL estimation.

Main Methods:

• 126 data pre-processing methods (statistical, wavelet transform, decomposition) were applied to input variables: observed GWL, evaporation, precipitation, and temperature.
• Four AI models were employed: Support Vector Machine (SVR), Artificial Neural Network (ANN), Long-Short Term Memory (LSTM), and Pelican Optimization Algorithm-ANN (POA-ANN).
• Akaike Information Criterion (AIC) was used to evaluate and validate the predictive accuracy of 1778 trained models.

Main Results:

• The average AIC values for ML, DL, and hybrid-ML classes decreased by -25.3%, -29.6%, and -57.8%, respectively, after applying pre-processing methods.
• Not all pre-processing methods improved prediction accuracy; careful selection through trial and error is essential.
• The wavelet-ANN model (db13_ANN_25) achieved the best GWL prediction with an AIC of -204.9, outperforming the non-pre-processed ANN model by 5.23%.

Conclusions:

• Data pre-processing significantly impacts the accuracy of AI models for groundwater level prediction.
• Hybrid-ML models, particularly the wavelet-ANN combination, demonstrate superior performance in GWL forecasting.
• The findings underscore the importance of judicious pre-processing method selection for effective water resource management.