Personality Type Prediction using Decision Tree, GBDT, and Cat Boost

Abstract

Personality scale analysis assists people in acquiring an accurate self-perception that facilitates their academic achievement and vocational success. The Big Five traits, as a combination of statistics and semantics, provide dimensional criteria to describe people’s characteristics and relevant behaviors with numerical estimation. However, a compositive evaluation can not be obtained by separately assessing this 5-factor model. To overcome the challenge, this paper uses decision trees, gradient boosting decision trees (GBDT), and cat boost to establish summative categorizations based on the five attributes. Firstly, the dataset is pre-processed to eliminate abnormal data while categorical data is converted into numerical form. Afterward, the dataset is analyzed in broken-lines and a heatmap to attest if the Big Five traits are unordered categorical variables. Finally, the dataset is trained in decision trees, GBDT, and cat boost to predict every examinee’s personality among five targeted types. To verify the effectiveness of the related methods, the accuracies of the three classifiers, which are differentiated by gradient constructions and algorithmic principles, are calculated and compared. The predictive accuracies of decision trees, GBDT, and cat boost are 0.52, 0.68, and 0.78, respectively. The results illustrate the feasibility of using classic algorithms in psychometric analysis, since personality types, based on the Big Five traits, are predictable with using decision trees, GBDT, and cat boost.