Ordinal Data Classification using Support Vector machines with Imbalance Data

Abstract

Ordinal data classification presents a complex challenge of training a model to correctly categorize observations within ranked data. However, real-world datasets utilized in ordinal classification frequently exhibit imbalanced class distributions, which pose a persistent obstacle in accurately classifying ranked data. The class imbalance issue often results in a bias toward majority classes within most classification models, leading to reduced accuracy and precision for minority classes. Additionally, successful implementations of classification models in finance sector with imbalanced classes remain limited.. Hence, this research paper introduces an innovative approach involving a hybrid class balancing technique followed by the utilization of Support Vector Machines (SVM) as the classifier for classifying mutual fund rating that are ordinal in nature. The study comprehensively compares the proposed hybrid SVM model and ordinal logistic regression in imbalanced data. Furthermore, the research extends its application to predicting mutual fund ratings and other relevant ordinal class data scenarios. Through empirical investigations conducted on both artificial and real-world datasets, including an application in Mutual fund rating analysis, this research establishes the efficacy and practical utility of the proposed approach.

Based on an empirical experiment, we assess the effectiveness of SVMs in identifying mutual fund ratings, along with implementation resampling methods typically utilized to tackle class imbalances.