Enhancing Anomaly Detection in Radio Astronomy Using Improved Isolation Forest Techniques

Project Abstract

The identification of anomalies stands as a critical analysis task in astronomy which helps scientists discover infrequent celestial occurrences and both natural transients and man-made instrumental artifacts. The research project focuses on enhancing the Isolation Forest (iForest) algorithm because of its fast operation and ability to analyze high-dimensional data. The iForest algorithm shows numerous weaknesses relative to other methods yet its specific weakness for this project involves both high sensitivity to hyperparameter tuning and poor performance with clustered anomalies. This research tackles the iForest algorithm deficiencies through enhanced feature selection methods and statistical heuristics and hybrid model development. The research project draws its data from the MeerKAT radio telescope which operates in South Africa as a leading instrument before the Square Kilometre Array (SKA) becomes operational. The MeerKAT telescope produces a massive daily quantity of astronomical data which has not been adequately analyzed. The research uses machine learning methods to analyze unused data sources which will improve both accuracy and robustness of radio astronomy anomaly detection. The proposed enhancements will enable scientists to make novel discoveries while advancing our knowledge of space.