Abstract

Groundwater is vital for public health, industry, and agriculture, and it is found under the surface in soil pores and rock fissures. Accurate modeling and prediction of groundwater parameters are required to ensure effective resource management and environmental sustainability. While the Gamma distribution is commonly used for forecasting groundwater features, it is limited to describing data with a right-skewed shape (where most values are low, but a few are very large). In this paper, we introduce the Odd Beta Prime Gamma (OBP-Gamma) distribution, a flexible statistical model that can describe both left- and right-skewed patterns as well as hazard rates (the probability that failure or contamination occurs at a given time). The OBP-Gamma distribution is applied to two groundwater parameters, pH and conductivity, and compared with classical Gamma and Weibull-Gamma models. Results show that OBP-Gamma provides a better fit for the observed data. In addition, we evaluated the use of machine learning models to classify groundwater potability using a small dataset of 30 water samples collected in Jaen, Kano State, Nigeria. Fourteen models were tested, and Gaussian Naive Bayes achieved the highest classification accuracy (90%), followed by Gradient Boosting (83.3%). Other models, such as Passive Aggressive and AdaBoost, performed poorly, with accuracy below 50%. These results highlight that the OBPGamma model offers improved flexibility for groundwater data analysis and that machine learning methods, particularly Gaussian Naive Bayes, show potential for assessing groundwater potability. However, due to the small sample size, the findings should be viewed as a proof-of-concept, with future research needed on larger datasets to confirm generalizability.