r0sh1ddy/Depression-Anxiety-predictor-for-High-School-Students-in-Kenya

This project aims at predicting anxiety and depression cases among high school students in Kenya. It spans about four counties in particular. Nairobi, Kiambu, Makueni, and Machakos. It aims to uncover the underlying factors contributing to depression and anxiety. It provides initiatives to be taken by the stakeholders to ensure immediate action. # Mental Health Screening - Kenya # AdolescentMind: Predicting Anxiety and Depression in Adolescents ## Project Overview This project develops an **Machine Learning Model** to predict depression (PHQ-8) and anxiety (GAD-7) symptoms among adolescents in Kenya. Using machine learning trained on local survey data, it powers an **interactive web app** that provides early mental health risk assessments aligned with Kenya’s *Mental Health (Amendment) Act, 2022*. > **Disclaimer:** This tool is for **screening**, not diagnosis. > For professional help, consult a qualified healthcare provider. ## Business Understanding Mental health challenges among adolescents are rising. The **WHO (2023)** reports that 1 in 7 teenagers (ages 10–19) experience a mental health disorder. This project aims to create a **data-driven early detection system** using responses from the PHQ-9 and GAD-7 questionnaires plus demographic information to identify students at risk of depression or anxiety. **Objectives:** - Build models that classify risk levels for depression and anxiety. - Improve early screening access in schools through an interpretable app. - Support national mental health policies focused on prevention and education. ## Methodology 1. **Data Preparation:** Cleaned, encoded, and scaled demographic and screening data. 2. **EDA:** Used `seaborn` and `matplotlib` to visualize distributions, correlations, and outliers. 3. **Modeling:** Trained Logistic Regression, Random Forest, XGBoost, and LightGBM models. 4. **Evaluation:** Measured performance using accuracy, recall, precision, and F1-score(Recal and Accuracy being the main metrics). 5. **Interpretability:** Applied **SHAP** to identify key influencing features. 6. **Deployment:** Integrated all our models into an interactive **Streamlit** application where the model with the best recall for each target is chosen in real-time to do a prediction. ## Key Insights - **Best Model:** LightGBM (strongest recall and interpretability). - ** …