MBTI Personality Prediction Project

This project was developed with the goal of automating MBTI personality type predictions using user-generated text.

Problem Statement

Traditional MBTI personality assessments rely on self-reports, which can be subjective and biased.
Goal: Predict personality types based on text data, offering a scalable and objective solution for:

Recruitment & candidate-job fit
Personalized recommendations
Mental health applications
Ad targeting & behavioral analysis

Workflow

1. Data Cleaning

Tokenization
Removal of URLs, special characters, stopwords

2. Feature Engineering

Traditional: TF-IDF
Deep Learning: Token Embeddings

3. Model Training

Compared Logistic Regression, Random Forest, and GRU (deep learning)

4. Evaluation Metrics

Accuracy, Precision, F1-score
Stratified train/test split to reduce class imbalance impact

5. Deployment Consideration

API-based prediction system for scalable real-time usage

Models & Performance

Model	Accuracy	Precision	F1-score
Logistic Regression	0.6406	0.66	0.64
Random Forest	0.6083	0.64	0.59
GRU (Deep Learning)	0.6486	0.7974	N/A

Logistic Regression: Fast, interpretable, but weak in semantic nuance
Random Forest: Great for short-texts and robust to overfitting
GRU: Best for long-form text like Reddit posts and job applications

Insights

GRU provided the best accuracy and is suited for real-world, high-stakes predictions
Logistic Regression still performs well in constrained environments
Wordcloud visualizations helped illustrate key lexical differences between personality types