Comparison of Explanation-centered Statistical Model and Prediction-centered Machine Learning
Sohee Koo, Seho Maeng, Sujin Park, Young‐Il Cho · The Journal of Humanities and Social sciences 21 · 2021
This paper compares the results of random forest and logistic regression analysis with empirical data. As the number of cases increased in logistic regression, the type 1 error and verification power were excessively high, leading to problems in predictive power. Although Random Forest showed higher predictive power, since all predictors were included in the analysis without theoretical background, the interpretation of the results were ambiguous or a variables that violate the Basic Law on Employment Policy were used. When analyzing big data, it is necessary to clean data appropriate for the purpose of the study. This paper proposes that we should be careful about model overfitting problem, and by using variable extraction or proxy variables, it can solve inequality while improving predictive power in the HRM.