Hybrid Chaotic Pattern Search Algorithm to Find Optimal Features for Water Potability and Hyperparameter Tuning for Random Forest Model

Water is essential for everyone to sustain life. It is used in various chores such as cleaning, drinking etc. Contaminated water can cause many diseases and hence potability of drinking water is significant. A novel approach for water potability classification has been proposed in our work through Hybrid Chaotic Pattern Search Algorithm (HCPSA) to find the best features and optimal hyperparameters for training Random Forest model to maximize accuracy. This algorithm effectively finds the global optimum by combining Chaotic stage, where Chebyshev chaotic map systematically explores the solution space and converge at a near optimal region, along with the Pattern search stage where exploratory and pattern moves are performed near the optimum solution obtained in the first stage. The accuracy of 91.49% demonstrates the effective handling of missing values and class imbalance handling via SMOTE and superiority of our algorithm in determining the best features and optimal hyperparameters.