Performance of Supervised Learning Algorithms on Imbalanced Class Datasets
Nur Anisah Binti Ramli, Maria Jasmin Binti Mohamed Jamil, Nur Nazifa Binti Zhamri, Mustafa Ali Abuzaraida · Journal of Physics Conference Series · 2021
Abstract In this paper, we measure the performance of supervised learning algorithms on imbalanced class datasets. Supervised learning is considered to be the most advanced and mature from other types of learning in machine learning. On the other hand, imbalanced class data sets refer to an unequal amount of data for each class in the data sets. Many real-world datasets exhibit an imbalanced class distribution. Hence, this paper aims to compare the supervised learning algorithms in classifying the output for imbalanced dataset. This paper also focused on finding the significance of balancing the dataset to the results. This study is conducted by using three different datasets and made a comparison with three supervised learning algorithms chosen which are Logistic Regression, Random Forest and k-Nearest Neighbours. In the experiments, all three datasets used are labelled data with imbalanced class. Due to the imbalanced class for all datasets chosen, a sampling technique with combination of under-sampling and oversampling is implemented to all datasets in data preparation steps. The performance of the algorithms is measured through Classification Accuracy (CA) value. From the experiments, it is proven that Random Forest shown the best results, for both imbalanced and balanced datasets.