Comparison of Machine Learning Models on Balanced and Imbalanced Dataset of Rodent Tuber

Abstrak

In this study, we investigate the performance of various machine learning models on both balanced and unbalanced datasets. The data processing pipeline begins with loading and preprocessing the dataset, followed by two parallel approaches: handling balanced and unbalanced data. For the balanced data approach, we employ random undersampling to achieve class balance. The Rodent Tuber's unbalanced dataset, which includes chemical data useful for cancer diagnosis, was used in this study. We use and evaluate Random Forest classifiers and Support Vector Machines (SVM) with three different kernels: sigmoid, linear, and radial basis function (RBF) on balanced and unbalanced datasets. Based on our research, the Random Forest algorithms generated the best classification results on balanced datasets, achieving a total 8.9second runtime with 99% accuracy, precision, recall, and$F 1$score. On balanced datasets, however, SVM performance differed greatly between kernels; linear kernels produced 50% accuracy, RBF kernels 61%, and sigmoid kernels 45%. The models' ineffectiveness is shown by their poor performance on unbalanced datasets, which mostly identified the majority class. This emphasize the negative impact of class imbalance. Furthermore, running the algorithm on the imbalanced dataset took significantly longer than on the balanced dataset.

Identifier

• OpenAlex: https://openalex.org/W7154572472
• DOI: https://doi.org/10.1109/icoabcd67551.2025.11470779