Detecting Groundwater Quality with Optimizing Gradient Boosting Performance: The Role of Focal Loss in Tree-Based Residual Models

Abstract

Globally, groundwater serves as a vital source for drinking water and agricultural purpose for billions of people. Assessing the quality is indispensable for sensing harmful contaminants, comprising nitrates, heavy metals and pathogens. The constant observing and evaluation are significant to decrease the risks associated to the ground water contamination and to provide safe drinking water for people. Manual methods for assessing groundwater quality involves physically collecting samples from boreholes or monitoring wells and exposing them to laboratory analysis. Though these manual methods provide comprehensive insights into quality, being resource-intensive and time-consuming by the frequency and number of sampling sites. To solve this issue, several researches have concentrated on water quality detection. Conversely, it results with lacks in accuracy and analysing in safety or not. For feature optimisation, PCA (Principal Component Analysis Optimization) Model is utilised. For enhancing the classification performance, the proposed method employs ML (Machine Learning) approach called Optimized Gradient Model with Effective analysis on trees with focal loss. It utilizes water quality dataset for evaluation. For improving accuracy in performance, the present research incorporates focal loss method which eradicates the imbalances in the dataset. Correspondingly, efficacy of the present model is calculated utilising various performance metrics are F1 Score, recall, precision and accuracy to estimate the performance. Further, the internal comparison of proposed models such as AdaBoost, XG Boost, Gradient Boosting and Random Forest (RF) and traditional models discloses the effectiveness of the present ML method. The projected research envisioned to contribute the emerging water quality detection models thereby, protecting public health and preserving safety environment.