XPredators Dementia Prediction

Overview

This project implements a machine learning pipeline for dementia prediction using a dataset hosted on Google Drive. The pipeline includes data loading, exploration, preprocessing, feature selection, dimensionality reduction via PCA, and training/evaluation of Logistic Regression and Random Forest models, ultimately selecting Logistic Regression for better generalization.

Key Features

• Data Loading: Downloads and loads a CSV dataset from Google Drive, consisting of approximately 195,000 rows and 1,024 columns with mixed data types.
• Data Preprocessing: Removes medical-related columns, converts data to numeric, fills missing values with medians, and applies variance thresholding to eliminate low-variance features.
• Feature Selection: Identifies and removes low-information columns (e.g., those with >80% single value frequency) and highly correlated features (>0.9 correlation).
• Dimensionality Reduction: Performs Principal Component Analysis (PCA) with 30 components after standardization to reduce dimensions while analyzing key loadings.
• Model Training and Evaluation: Trains Logistic Regression and Random Forest models on a 60/40 train/test split, evaluates using accuracy, precision, recall, F1-score, and confusion matrices; compares performance to select the best model.

Tech Stack

• Python Libraries:
  • gdown for Google Drive downloads
  • pandas and numpy for data manipulation
  • scikit-learn for preprocessing (VarianceThreshold, StandardScaler), PCA, LogisticRegression, RandomForestClassifier, and metrics (accuracy_score, classification_report, confusion_matrix)
  • matplotlib and seaborn for visualization (heatmaps, confusion matrices)

Project Structure

• .gitignore: Configuration file specifying patterns for files to ignore in version control (e.g., *.docx, *.py, *.tmp).
• .ipynb_checkpoints/Untitled-checkpoint.ipynb: Automatically generated checkpoint for notebook state recovery (empty JSON structure).
• XPredators_Demantia_Prediction.ipynb: Main Jupyter notebook containing the complete dementia prediction pipeline, from data download to model evaluation.

Setup Instructions

1. Ensure Python 3.x is installed on your system.
2. Install required dependencies using pip:

   pip install gdown pandas numpy scikit-learn matplotlib seaborn
   

3. Open the project in a Jupyter Notebook environment (recommended: Google Colab).
4. Upload XPredators_Demantia_Prediction.ipynb or clone the repository.
5. Ensure internet access for Google Drive downloads if running locally.

Usage

1. Run the notebook cells sequentially starting from the first cell.
2. The notebook will:
   • Download the dataset using gdown with file ID 19mKGPNFb35kG__3Eihazyv5O69ZUxDcF.
   • Perform exploratory data analysis, preprocessing, and feature selection.
   • Apply PCA for dimensionality reduction.
   • Train Logistic Regression and Random Forest models.
   • Display evaluation metrics, confusion matrices, and plots.
3. View outputs such as processed DataFrames (X_selected, y), trained models (lr_model, rf_model), and performance reports.

Key configurations include:

• VarianceThreshold: threshold=1
• Low-information removal: >80% single value frequency
• Correlation threshold: >0.9 for feature dropping
• PCA: n_components=30
• Random Forest: n_estimators=300, max_depth=None, random_state=42
• Logistic Regression: penalty='l2' (default)
• Train/test split: test_size=0.4, random_state=42

Contributing

Contributions are welcome. Please submit pull requests with clear descriptions of changes or open issues for feature requests and bug reports.

License

This project is licensed under the MIT License. See the LICENSE file for details.