Personality Prediction (Kaggle S5E7)

This repository contains the solution for the Kaggle Playground Series - Season 5, Episode 7 competition. The goal is to predict a person's 16-type personality based on their answers to a custom survey.

This project uses an XGBoost Classifier to build four independent binary models—one for each of the four personality dichotomies—to achieve a robust and accurate prediction.

---

🚀 Project Overview

The 16 personality types (like INTJ, ESFP, etc.) are a combination of four binary traits:

1. Introversion (I) vs. Extroversion (E)
2. Intuition (N) vs. Sensing (S)
3. Feeling (F) vs. Thinking (T)
4. Perceiving (P) vs. Judging (J)

Instead of treating this as a complex 16-class classification problem, this solution builds four separate binary classification models. The final personality type is determined by concatenating the results of these four models (e.g., I + N + T + J = INTJ).

---

🔧 Methodology & Pipeline

1. Data Preprocessing

• Loading: The train.csv and test.csv files are loaded using pandas.
• Target Engineering: The single Personality column (e.g., 'INFP') in the training data is split into four separate binary target variables: is_I, is_N, is_F, and is_P.
  • 'INFP' -> is_I=1, is_N=1, is_F=1, is_P=1
  • 'ESTJ' -> is_I=0, is_N=0, is_F=0, is_P=0
• Features: The survey questions serve as the features (X). The id column is dropped.

2. Modeling with XGBoost

• An XGBClassifier is used as the base model for its high performance and speed.
• Four models are trained independently:
  1. model_IE: Predicts is_I (Introversion/Extroversion) using X
  2. model_NS: Predicts is_N (Intuition/Sensing) using X
  3. model_FT: Predicts is_F (Feeling/Thinking) using X
  4. model_PJ: Predicts is_P (Perceiving/Judging) using X
• The Modeling.ipynb notebook establishes this baseline approach, while Modeling2.ipynb likely focuses on hyperparameter tuning (e.g., GridSearchCV or RandomizedSearchCV) to optimize each of the four models.

3. Prediction & Submission

1. The test data (test.csv) is loaded.
2. Each of the four trained models (model_IE, model_NS, etc.) predicts its respective binary class on the test data.
3. The four binary predictions are mapped back to their letter codes (e.g., 1 -> 'I', 0 -> 'E').
4. The final Personality string is created by concatenating the four predicted letters.
5. The results are formatted into submission.csv with id and Personality columns.

📈 Hyperparameter Tuning & Results

The models were tuned (likely using Bayesian optimization or a similar method, as seen in Modeling2.ipynb) to find the optimal set of parameters.

Best Parameters Found

{
    "n_estimators": 299,
    "max_depth": 5,
    "learning_rate": 0.15841262137302178,
    "subsample": 0.8519152889164038,
    "colsample_bytree": 0.6808885474211932,
    "gamma": 2.0070959113867732,
    "reg_alpha": 1.2522715414146957,
    "reg_lambda": 2.5895571241033593
}

Performance Metrics

• Best CV Score: 0.9639316464361883
• Tuned Model R2 (Train): 0.9640534903692799
• Tuned Model RMSE: 0.30210824789781193

---

💻 How to Run

1. Clone the repository:

   git clone [https://github.com/RaymussenArthur/Personality-Prediction.git](https://github.com/RaymussenArthur/Personality-Prediction.git)
   cd Personality-Prediction

2. Install dependencies: It's recommended to use a virtual environment.

   pip install pandas numpy scikit-learn xgboost jupyter

3. Get the data:

   • Download the train.csv, test.csv, and sample_submission.csv files from the Kaggle competition page.
   • Place them inside the /data folder.

4. Run the notebooks:

   • Start Jupyter:

     jupyter notebook

   • Open and run the notebooks in the Notebooks/ directory, starting with Modeling.ipynb and Modeling2.ipynb to train models and generate a submission.