← Back to Projects
Completed

Jakarta Air Quality Index (ISPU) Prediction

Predicting Jakarta's air quality index using 5 ML algorithms with SHAP explainability.

Best accuracy: 94.2% (Random Forest)
PythonScikit-learnSHAPPandasMatplotlib

2024-11-15

Background

Jakarta consistently ranks among the cities with the worst air quality in the world. Accurate ISPU (air quality index) prediction helps residents and the government make timely decisions.

Approach

Compared 5 algorithms: Logistic Regression, Decision Tree, Random Forest, SVM, and KNN. Used SHAP to explain why the model makes a given prediction, not just how accurate it is.

Key Findings

  • Random Forest achieved 94.2% accuracy, consistently outperforming the other models
  • PM2.5 was the feature with the largest influence on predictions
  • The model tends to underpredict at the “Very Unhealthy” level, likely due to imbalanced data

What I Learned

High accuracy isn’t enough. A model that can’t be explained is hard to trust in a public health domain. SHAP changed how I think about model validation.