Incremental learning using CART and Genetic Programming for handling dynamic training data
Ibadonbok Syiemlieh, Ymphaidien Sutong, Sufal Das · 2025
Genetic Programming (GP) is a technique that focuses on evolving a program from a population of unfit programs to fit a specific task by applying operations similar to natural genetic processes on the program population. GP can be used to dynamically adapt tree classifiers to concept drift in data streams. The sudden changes in data may lead to concept drift, which is one of the significant challenges in machine learning. This paper presents a new technique for classification with dynamic data through the hybridisation of GP and the CART algorithm. CART is a well-established decision-tree-based classifier. This proposed method efficiently handles concept drift, in which the pattern of data changes over time and affects the model’s performance. The proposed approach improves decision tree adaptability using Selection, Crossover, and Mutation to iteratively make model populations. Several experiments were conducted with breakneck-mark datasets. We have also compared the developed decision tree to baselines models and other methods. The paper helps deepen the understanding of adaptive machine learning approaches in dynamic data, where continuous learning is essential. Metrics such as accuracy, precision, recall, and F1-score are used for evaluating performance.