Machine Learning Models With Minimum Dataset in Text-Based Personality Prediction

There has been extensive research in the field of text-based personality prediction. However, one of the primary challenges in text-based personality prediction lies in the issue of imbalanced data and the difficulty of achieving accurate predictions when training data is limited. This research investigates the use of machine learning for predicting Myers-Briggs Type Indicator (MBTI) personality types based on social media text data. The study employs multiple models, including Gradient Boosting, CatBoost, XGBoost, and GPT to classify texts into 16 distinct MBTI categories. Results reveal that the XGBoost model achieves the highest accuracy, reaching 89.06 % on a large dataset comprising over 100,000 data points, significantly outperforming results obtained on smaller datasets, but using GPT can achived 86.6 % using small dataset (180 data points). This study contributes to the development of personality prediction tools using small datasets.