🚀 Addiction Risk Prediction

This repository provides a comprehensive machine learning pipeline designed to predict addiction risk using behavioral, psychological, and demographic data. The solution empowers early identification of high-risk individuals and supports data-driven intervention strategies.

---

🎯 Project Objectives

The goal of this project is to predict the likelihood of addiction (Yes/No) based on user-specific inputs including:

• Substance usage (Alcohol, Cannabis, Tobacco, etc.)
• Age of first use and current age
• Frequency of usage
• Reported stress levels and diagnosed mental health conditions
• Coping mechanisms and presence of support systems

Through this, we aim to:

• Understand patterns of substance exposure, especially in adolescence or pre-teen years
• Explore behavioral risk factors contributing to addiction
• Enable predictive modeling to assist mental health professionals

---

📊 Dataset Overview

The dataset was collected firsthand using Google Forms, ensuring authentic, real-world insights into individual behavioral patterns.

• 📎 Google Form Link (Original Survey)

📁 Files Provided

• Dataset.xlsx – Raw collected data
• Cleaned_Encoded_Dataset.xlsx – Preprocessed dataset used for modeling
• New.xlsx – Data used for inference or testing predictions

---

⚙️ How to Run the Project

✅ Step-by-Step Instructions

1. Launch the Notebook

   • Open Addiction_Risk_Prediction.ipynb
   • Click “Open in Colab” for browser-based execution

2. Upload the Required Files

   • Download and upload the following to Colab:
     • Dataset.xlsx
     • Cleaned_Encoded_Dataset.xlsx
     • New.xlsx

3. Run the Notebook

   • Execute each cell sequentially to perform:
     • Data preprocessing
     • Feature engineering
     • Model training and evaluation
     • Final binary prediction (Addiction: Yes / No)

---

🤖 Machine Learning Approach

The project initially uses multiclass classification on the substances_used variable, which categorizes different types of substances. Post-training, the focus shifts to a binary classification task to determine whether an individual is at risk of addiction.

🔍 Models Benchmarked

• Random Forest ✅ (Best Performing Model)
• Support Vector Machine (SVM)
• XGBoost
• Logistic Regression
• K-Nearest Neighbors
• Decision Tree
• Naive Bayes

🧠 Ensemble Model

We also developed a stacked ensemble model to boost performance:

• Base Models: Random Forest, SVM, XGBoost
• Meta-Learner: Logistic Regression

Performance was evaluated using:

• Accuracy
• Precision
• Recall
• F1-score

---

📌 Real-World Insight

Our findings suggest that:

• Individuals are often exposed to addictive substances during pre-teen and adolescent years
• Stress levels, inadequate coping mechanisms, and mental health diagnoses are significant predictors
• Early intervention can be guided using such predictive models

---

📈 Future Roadmap

• Deploy as a REST API or interactive dashboard
• Integrate interpretability tools (e.g., SHAP, LIME)
• Automate data ingestion pipelines for real-time risk prediction
• Extend dataset for temporal analysis or time series modeling

---

📬 Contact

For contributions, queries, or to collaborate on health-tech initiatives, reach out via GitHub Issues.

---

📎 License

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

---

Developed with an intent to support mental health awareness, data literacy, and predictive analytics in public health.