Forecasting water quality is important for managing freshwater, especially where it supports farming and everyday life. This project compares traditional time series methods like ARIMA and SARIMAX with newer machine learning models, including XGBoost, LightGBM, and CatBoost, to predict groundwater quality in Nadia district, West Bengal, India, across four water quality monitoring stations. The study focuses on pH, Total Dissolved Solids (TDS), Electrical Conductivity (EC), and Sodium (Na) as key parameters. The process begins with careful data cleaning, handling missing values, removing outliers, performing feature selection using correlation analysis, and considering seasonality to ensure more accurate results. Model performance is evaluated using R², MAPE, and NRMSE scores. The findings show that, particularly after selecting the most important input features, the gradient boosting models outperform traditional statistical ones in capturing complex trends and providing more reliable predictions. This work highlights the role of modern data-driven forecasting methods in advancing data science applications for water quality assessment. It demonstrates how such approaches can support environmental sustainability by enabling better decisions on water use for farming and environmental management.