Assessment of weight factor in genetic programming fitness function for imbalanced data classification

Abstract

In real-world data classification, applications often have an imbalanced distribution of data over various classes. This imbalanced distribution imposes intense challenges, and because of this, traditional classification methods are not effective in this case. This problem also influences genetic programming (GP). One approach to resolve this issue is to assign a custom high weight to the classes during training. This custom weight assignment may nullify the impact of higher counts of any classes during model learning. In GP, this custom weight assignment may be introduced in the fitness function. The fitness function performs an essential role in GP and influences each building block of GP. In this research work, we assess the impact of weight factors on GP's fitness function for imbalanced data classification. For this assessment, eight imbalanced classification problems are taken from the UCI repository, and intensive experimentation is done on the different weight factors.