Machine Learning in Planetary Defense: A Random-Forest-Based Early Warning Framework for Hazardous Near-Earth Asteroid Detection

Manikonda Ramya Krishna1, Rodda Kumari2

1Department of Computer Science and Engineering, Vikas College of Engineering and Technology, Vijayawada, India.

2 Department of Computer Science and Engineering, Vikas College of Engineering and Technology, Vijayawada, India.

Abstract - Planetary defense is an emerging area of scientific and strategic importance that focuses on detecting and mitigating the threat posed by hazardous Near-Earth Objects (NEOs). Conventional survey pipelines depend heavily on human-led observation and classical orbital-mechanics analysis, an approach that is time-consuming, difficult to scale, and increasingly overwhelmed by the volume of data produced by modern sky surveys. This paper presents a machine-learning-based early warning framework that automates the detection, classification, and threat assessment of potentially hazardous asteroids (PHAs). The proposed pipeline aggregates orbital and physical parameters from public astronomical archives, engineers risk-relevant features such as the Minimum Orbit Intersection Distance (MOID), relative velocity, and orbital eccentricity, and trains a Random Forest classifier to label objects as hazardous or non-hazardous. The trained model is deployed through a lightweight Flask web interface that allows a user to input orbital parameters and receive an instantaneous hazard prediction. On a held-out test set, the system achieved an overall accuracy of 53% and a recall of 61% for the hazardous class, indicating that, despite a modest overall accuracy, the model favours minimizing false negatives — an outcome that is operationally desirable in planetary defense, where failing to flag a genuinely hazardous object is more costly than issuing an unnecessary alert. The paper further outlines a broader system architecture encompassing time-series forecasting, Monte-Carlo based impact-probability estimation, tiered alert generation, and continuous model retraining, and discusses directions for improving accuracy through dataset balancing, ensemble learning, and explainable-AI techniques.

Key Words:  Planetary Defense; Near-Earth Objects; Potentially Hazardous Asteroids; Machine Learning; Random Forest; MOID; Flask; Early Warning System.