Four personality traits based on MBTI, which are Introversion or Extroversion, iNtuition or Sensing, Feeling or Thinking, and Judging or Perceiving, are predicted from posts using Machine Learning algorithms such as, Logistic Regression, Random Forest, Naive Bayes, XG-Boost, and BERT The experiments make use of a Kaggle benchmark dataset that is openly accessible. The key problem with the earlier work is the dataset's skewness, which is reduced by using balancing approaches for better performance. Pre-processing and feature selection are used to explore the personality of the text. Based on questionnaire responses, this work offers a web application for personality identification.
All of the classifiers produced adequate results for all of the personality types, however combining the balancing techniques with the standard machine learning algorithms performs significantly better than traditional algorithms.
Out of all the algorithms implemented in this project we find that BERT model can understand the relevance and meaning of each word, whereas the other proposed models try to decrease error and fit the data without understanding the language, making BERT our best model for this project.
-
Perform data pre-processing and exploratory data analysis
-
Design and Implement classification model to predict the personality type.
-
Compare and validate the classifiers to identify the best classifier.
-
Develop a web application to take a new data as an input and displays the predictions as visualization.