Personality Type Prediction Project (Kaggle Playground Series - S5E7)

Project Overview

This project was developed for participation in the Kaggle Playground Series - Season 5, Episode 7 machine learning competition.
The main objective is to predict whether an individual is an Introvert or Extrovert based on their behavioral, social, and activity-related traits.

Dataset

The dataset contains survey-based features reflecting user behavior, including:

• Time_spent_Alone
• Stage_fear
• Social_event_attendance
• Going_outside
• Drained_after_socializing
• Friends_circle_size
• Post_frequency
• Personality (Target: Extrovert or Introvert)

Project Structure

The project mainly consists of:

• main.py: A python file containing all stages from data loading to submission, including EDA, feature engineering, model selection, and prediction.
• preprocessing.py: A reusable module that includes the data transformation pipeline for both train and test datasets. It ensures consistency between model training and final predictions.

Workflow

The pipeline includes the following steps:

1. Data Loading: Train and test datasets are loaded from Kaggle inputs.
2. Exploratory Data Analysis (EDA): Summary statistics, missing value inspection, and visualization of feature distributions.
3. Missing Value Imputation:
   • KNNImputer is used for numeric columns.
   • SimpleImputer with mode strategy is used for categorical variables.
4. Outlier Handling:
   • Outliers are capped using the IQR method (adjusted thresholds).
5. Feature Engineering:
   • New features like NEW_Alone_Level, NEW_Social_Score, and categorical bins were created to better capture personality traits.
6. Encoding:
   • Binary features were label encoded.
   • Multiclass categorical variables were one-hot encoded using sklearn.OneHotEncoder with handle_unknown='ignore'.
7. Model Comparison:
   • Several classifiers (CatBoost, XGBoost, LightGBM, SVC, RandomForest, etc.) were compared.
   • CatBoost (GPU) performed the best on the validation set.
8. Hyperparameter Optimization:
   • Optuna was used to tune CatBoost hyperparameters with cross-validation using f1_macro as the objective.
9. Final Prediction and Submission:
   • The final model was trained on the full training dataset using the best parameters.
   • Predictions were made on the test dataset and saved as submission.csv.

Setup and Running

To run this project on your local machine or cloud:

1. Clone the Repository:

   git clone https://github.com/BahriDogru/Personality_Type_Classification.git
   cd predicting_personality_type

2. Install Dependencies:

   Using the provided environment.yaml file:

   conda env create -f environment.yaml
   conda activate personality_prediction_env

3. Prepare the Dataset: Download train.csv and test.csv from the competition page
   Place them in a dataset/ folder:

   .
   ├── dataset/
   │   ├── train.csv
   │   └── test.csv
   ├── main.ipynb
   ├── environment.yaml
   ├── .gitignore
   ├── preprocessing.py
   └── README.md
   

4. Run the Script:

   python main.py

Results

After model comparison and tuning, the CatBoostClassifier (GPU) model provided the best results.
📈 Private Leaderboard Score: 0.974089 (F1 Score)