Ensuring access to safe drinking water is a fundamental public health priority. Traditional methods for assessing water quality are laborintensive and require specialized equipment, which may not be feasible for continuous monitoring. This study explores the use of machine learning models to predict water potability based on various chemical properties. Specifically, we evaluate the performance of Logistic Regression and Random Forest models in the presence of class imbalance, a common issue in environmental datasets. To mitigate this, we apply the Synthetic Minority Over-sampling Technique (SMOTE). Our results indicate that before applying SMOTE, both models exhibited a strong bias towards the majority class (non-potable water), achieving an accuracy of 69.36% and a ROCAUC of 0.63. However, the application of SMOTE significantly improved the model’s ability to identify potable water samples, particularly for the Random Forest model, which achieved an accuracy of 67.07% and a ROCAUC of 0.64 post-SMOTE. In contrast, the Logistic Regression model showed a decline in performance after SMOTE, suggesting the need for further optimization or alternative approaches. This study highlights the importance of addressing class imbalance in machine learning tasks, especially for critical applications like water quality assessment. Our findings suggest that the Random Forest model, combined with SMOTE, offers a robust solution for predicting water potability. These insights can aid environmental scientists and public health officials in implementing more efficient and accurate water quality monitoring systems. Future research should explore a broader range of models and advanced techniques to further enhance prediction accuracy.