Enhancing Classification Through Multi-view Synthesis in Multi-Population Ensemble Genetic Programming
Mohammad Sadegh Khorshidi, Navid Yazdanjue, Hassan Gharoun, Danial Yazdani, Mohammad Reza Nikoo, Fang Chen, Amir H. Gandomi · Proceedings of the Genetic and Evolutionary Computation Conference Companion · 2024
This study proposes a genetic programming (GP) approach for classification, integrating cooperative co-evolution with multi-view synthesis. Addressing the challenges of high-dimensional data, we enhance GP by distributing features across multiple populations, each evolving concurrently and cooperatively. Akin to multi-view ensemble learning, the segmentation of the feature set improves classifier performance by processing disparate data "views". Individuals comprise multiple genes, with a SoftMax function synthesizing gene outputs. An ensemble method combines decisions across individuals from different populations, augmenting classification accuracy and robustness. Instead of exploring the entire search space, this ensemble approach divides the search space to multiple smaller subspaces that are easier to explore and ensures that each population specializes in different aspects of the problem space. Empirical tests on multiple datasets show that the classifier obtained from proposed approach outperforms the one obtained from a single-population GP executed for the entire feature set.