A Modified GP based Feature Selection Model for Solving Real World Classification Problems

In this data-driven era, machine learning models face a critical hurdle: managing vast amounts of complex, high-dimensional data. These datasets, while rich in potential insights, often lead to slower computations, overfitted models, and opaque decision-making processes. To tackle these challenges, this study introduces a fresh approach using Genetic Programming (GP)—a technique inspired by natural evolution—to identify the most impactful features in datasets while boosting model performance.Breaking away from traditional feature selection methods, our framework combines dynamic feature weighting, diversity-driven population generation, and adaptive mutation strategies to iteratively refine feature subsets. Over 13 carefully designed steps, the method evolves classifiers across generations, prioritizing accuracy and simplicity. We tested this approach on eight different datasets from diverse fields like finance, healthcare, and education. The results shows that Taiwanese Bankruptcy Prediction dataset shows highest mean accuracy of 97.18% with selected features. It was also noted that both the dataset of educational field didn’t perform well, they scored the least ac-curacy score among all the datasets. Beyond numbers, this work bridges theory and practice, offering a scalable tool for researchers and practitioners to build faster, clearer, and more reliable AI systems.