Outlier detection is pivotal in data analysis, particularly with the ever-increasing dimensionality of datasets, which introduces the challenge of the "curse of dimensionality". The traditional Local Outlier Factor (LOF) algorithm, though effective in lower-dimensional spaces, struggles with high-dimensional data. In this paper, we propose an innovative approach, the InfoPrincipal Local Outlier Factor (IP-LOF), which is an enhanced method by integrating Mutual Information and Principal Component Analysis for improved outlier detection in high-dimensional spaces. IP-LOF processes data through dual pathways, applying LOF to subsets identified by these two methods, enabling a nuanced data analysis. Evaluations on synthetic and real-world datasets demonstrate IP-LOF's superior performance over LOF and other benchmark algorithms, particularly in terms of the Area Under the Receiver Operating Characteristic Curve (AUC). Our method illustrates robust adaptability and precision in outlier detection across diverse datasets, addressing the challenges posed by high-dimensional data while ensuring computational efficiency.
