Use of Exponential Transformations for Anomaly Detection

Despite the rapid growth in the use of deep learning for anomaly detection, in many cases these models achieve only modest improvements over classical shallow methods, while requiring significantly higher resources. Considering the competitive potential of lightweight methods, this paper introduces a novel unsupervised shallow method for anomaly detection. The presented method utilizes exponential transformations – low-cost operations, to emphasize deviations in data values, thereby making anomalies more separable. Experimental results demonstrated that our method outperforms existing shallow methods. Tested on larger datasets, the method achieved results similar to those of deep learning models, but with substantially lower computational cost and faster execution.