Ensemble Learning Paradigms for Flow Rate Prediction Boosting

In response to the issue of water scarcity in recent years, international organizations, in collaboration with many governments, have initiated several drinking water supply projects carried out by geophysical and drilling companies. Unfortunately, despite the reliability of electrical resistivity profiling (ERP) and vertical electrical sounding (VES) methods, the substantial financial losses incurred due to numerous unsuccessful drillings are owing to the difficulty to emphasize the drilling location properly. Therefore, we proposed the ensemble machine learning (EML) paradigms to predict the flow rate (FR) with an optimal score before any drilling operations. The approach was experimented in a region with severe water shortages. Thus, geo-electrical features from the ERP and VES were defined and coupled with borehole data to create the binary dataset $$( FR\le 1{m}^{3}/hr$$ ( F R ≤ 1 m 3 / h r and $$FR>1 {m}^{3}/hr$$ F R > 1 m 3 / h r for unproductive and productive boreholes respectively). Then, the dataset is state-of-art transformed before feeding to the EML algorithms. The model performance and generalization capability were evaluated using the Matthews correlation, the accuracy, the confusion matrix, the binary predictor error, the precision-recall, and the cumulative gain plot. As a result, the benchmark, pasting, extreme gradient boosting, and stacking paradigms have built a powerful range of FR prediction scores between 90 ~ 96%. Henceforth, the robust EML paradigms can be used to identify the best location for drilling operations, lowering the repercussion of unsuccessful drillings.