Enhancing Machine Learning Performance on Imbalanced Resume Data Using SMOTE for Job Candidate Classification

Abstract

The problem of data imbalance in machine learning processes is a significant limitation affecting model performance, particularly when the minority class has considerably fewer samples than the majority class. This imbalance causes the model to be biased and less accurate in classification. A widely adopted solution to this issue is using SMOTE, which generates new synthetic samples for the minority class by calculating the distance between existing data points. This research presents the application of SMOTE to improve the performance of machine learning models, specifically Decision Trees, Random Forests, and K-NN, on imbalanced datasets. Results indicate that the accuracy of the models improved from 0.83, 0.86, and 0.80 to 0.84, 0.88, and 0.82, respectively; there is statistical significance after applying SMOTE. This demonstrates the effectiveness of the technique in addressing data imbalance issues. The findings confirm that SMOTE enhances the ability to classify minority class data, resulting in models that are more accurate and suitable for applications in contexts where data imbalance is a concern. By helping reduce the time spent on personnel selection for the recruitment department.