Empirical Analysis using Feature Selection and Bootstrap Data for Small Sample Size Problems
Yuying Zhao, Rakkrit Duangsoithong · 2019
With the emergence of more application scenarios, the classification accuracy of small sample size problems especially in some areas such as medical data, needs to be further improved. This paper presents an empirical analysis using feature selection and bootstrap data to overcome this problem. Four traditional classification models were used in the experiment: k-Nearest Neighbors, Naive Bayes, Decision Tree and XGBoost. According to the result, the bootstrap method especially with XGBoost classifier provides better classification accuracy than using feature selection method and the data that randomly selected from original data, respectively.