Machine Learning · Classification · NASA Open Dataset
Hazardous Asteroid Detection Model
Classifying near-Earth objects as hazardous or non-hazardous where the cost of a false negative is not symmetric with a false positive.
Overview
A binary classification model on NASA's near-Earth object dataset that prioritizes recall, because in hazard detection a missed positive costs far more than a false alarm.
Data & inputs
NASA's open near-Earth object dataset, covering orbital and physical characteristics of catalogued objects.
Approach
Preprocessing and feature engineering followed by StandardScaler normalization, with Logistic Regression as an interpretable classifier whose coefficients can be inspected directly.
Evaluation
Recall was chosen as the primary metric because a missed hazardous object is the expensive error; precision, F1-score and a confusion matrix were used alongside it to keep the false-positive cost visible.
- Recall (primary)
- Precision
- F1-Score
- Confusion matrix
Key learnings
Metric choice is a modelling decision, not a reporting one. Selecting recall as the objective changed how thresholds were evaluated and made the accuracy figure largely irrelevant to the problem.