Effective Hazard Categorization of Near-Earth Objects with Random Forest Techniques
Rekha R Nair, Tina Babu
Abstract
Near-Earth Objects (NEOs) have long been a subject of interest for astronomers and planetary defense experts due to their potential impact risk. Hence, Random Forest techniques were utilized to classify NEOs as hazardous or non-hazardous using NASA’s comprehensive dataset of 338,199 NEO observations from NASA, spanning from 1910 to 2024. The data underwent preprocessing, including standard scaling and the removal of outliers. Various visualization methods were applied to elucidate patterns within the dataset. The Random Forest model achieved a 95% accuracy rate in distinguishing between hazardous and non-hazardous NEOs after outlier removal. The model demonstrated precision values of 0.95 and 0.70 for non-hazardous and hazardous classes respectively, with corresponding recall values of 0.97 and 0.58. This research underscores the efficacy of Random Forest algorithms in assessing NEO hazards and discusses the implications for prioritizing observational resources and enhancing early warning systems for potential NEO threats.