Hazardous Asteroids Detection

This project analyzes Near-Earth Object (NEO) data to detect potentially hazardous asteroids using three classification models:

• Logistic Regression
• Random Forest
• XGBoost

Objective

The goal is to assess the effectiveness of various supervised machine learning methods in identifying asteroids that may pose a risk to Earth. The models focus on maximizing recall to reduce the chance of missing dangerous objects.

Dataset

Source: NASA NEO dataset (assumed origin – can be adjusted)

Features Used

• Absolute magnitude
• Estimated minimum diameter
• Estimated maximum diameter
• Relative velocity
• Miss distance

Models & Evaluation

Three models were trained and tested with class imbalance taken into account:

• Logistic Regression: good recall (93%) but low precision
• Random Forest: improved accuracy, limited gain in recall
• XGBoost: best recall (93%), but still low F1-score due to false positives

Visualizations include:

• Confusion matrix
• Top 10 most dangerous asteroids (by predicted probability)

Technologies Used

• Python (pandas, scikit-learn, xgboost, seaborn, matplotlib)
• Jupyter Notebook

Key Insights

• Recall is prioritized over precision in this case to ensure dangerous asteroids are not missed.
• XGBoost provided the most reliable detection rate for hazardous objects.
• Feature scaling and class weighting significantly impacted model performance.

Final Output

Bar chart of the top 10 most hazardous asteroids with predicted probabilities (based on XGBoost model).

---