Classification of Potentially Hazardous Asteroids Using Artificial Neural Networks and Over Sampling Techniques

Addressing the challenge of imbalanced datasets is a common concern in machine learning, particularly when dealing with rare events or outcomes. The issue arises when there are insufficient samples representing minority classes, resulting in biased models and inaccurate predictions. In this research project, we tackle the classification of potentially hazardous asteroids (PHAs), a task exacerbated by the imbalanced nature of the dataset. Our study utilizes the PHA dataset generously provided by NASA's Jet Propulsion Laboratory (JPL), containing crucial information regarding the orbital and physical characteristics of PHAs. To counter the class imbalance and enhance classification accuracy, we propose a novel solution employing oversampling techniques. Specifically, we leverage Synthetic Minority Over-sampling Technique (SMOTE) and its advanced counterpart, Borderline SMOTE, in conjunction with an Artificial Neural Network (ANN) model for classification. By applying these techniques, we effectively balance the dataset, ensuring more representative samples for the minority class. This equilibrium leads to substantial improvements in the accuracy of PHA identification and a notable reduction in the risk of false negatives. Notably, in our results, we evaluate the performance using key metrics such as the F1 score and precision. It is worth highlighting that Borderline SMOTE outperformed SMOTE significantly, demonstrating its superior efficacy in handling the imbalanced data.