Student Depression Prediction Model

Note: You can access the repo here.

Depression among college students is a significant and growing concern. According to the Healthy Minds Study (2022), 44% of college students report symptoms of depression, reflecting a steady rise in mental health challenges over the past decade. The National College Health Assessment (2023) further reveals that 30% of students seriously consider seeking help, underscoring the need for timely intervention.

prediction demo video

Task

My objective was to develop a machine-learning model capable of predicting the likelihood of depression among college students. This involved analyzing various factors such as academic performance, social interactions, and lifestyle habits to identify key predictors of depression.

Action

1. Data Collection and Preprocessing:

   • The dataset used in this analysis was obtained from Kaggle.

   • Addressed missing values and normalized data to ensure consistency and accuracy.

2. Exploratory Data Analysis (EDA):

   • Conducted exploratory analyses to identify correlations between potential predictors and depression symptoms.

   
   

   

   Features Correlation with Depression

   • Visualized data distributions and relationships to inform feature selection.

     
     

     

     
     

3. Feature Engineering:

   • Selected relevant features such as Age, CGPA, academic pressure, study satisfaction, and others.

4. Model Development:

   • Tested various machine learning algorithms, including logistic regression, random forest, decision tree, and XGBoost.

   • Performed hyperparameter tuning to optimize model performance.

Turning Random Forest max_depth value

5. Model Evaluation:

• Assessed models using metrics such as accuracy, precision, recall, and F1-score.

  
  

  

  models accuracies, AUCs scores...

  6. Deployment and Documentation:

• Deployed the final model as a user-friendly application for institutional use.

• Documented the entire process, including code, methodologies, and findings, in a comprehensive GitHub repository.

Result

• Model Performance:

  • Achieved an accuracy of 85%, precision of 85%, and recall of 87%, indicating robust predictive capabilities.

• Key Predictors Identified:

  • High academic stress and financial stress were strongly associated with increased depression risk.

  • Positive attributional styles, where students credit successes to their own efforts and abilities, correlated with lower depression susceptibility.

• Impact:

  • Provided a scalable tool for educational institutions to identify and support at-risk students proactively.

  • Contributed to the understanding of factors influencing student mental health, aiding in the development of targeted interventions.

    
    

Thanks for coming this far!