Anomaly Detection and Performance Analysis With Exponential Smoothing Model Powered by Genetic Algorithms and Meta Optimization

Abstract

This study employs a genetic algorithm to optimize the parameters of the Third Order Exponential Smoothing model for predicting on the real-time traffic datasets of the Numenta Anomaly Benchmark (NAB). The genetic algorithm process was executed with different population sizes and gene sets. In addition, a parameter sensitivity analysis was conducted, through which the ideal number of genes and population size providing the best results within the specified range were determined. Moreover, a novel approach incorporating meta-optimization techniques is proposed to enhance the efficiency of the genetic algorithm optimization process, aiming to achieve improved accuracy in anomaly detection. The proposed methodology has been tested on various traffic data scenarios across different datasets to detect deviations critical to traffic management systems. Performance comparisons using the NAB scoring system demonstrate that the method developed in this study outperforms the majority of existing NAB algorithms, as well as the contemporary approaches of Isolation Forest, Multi-Layer Perceptron Regressor (MLPRegressor), and hybrid K-Nearest Neighbors - Gaussian Mixture Models (KNN + GMM), and is competitive with leading algorithms. The proposed approach, which achieved scores of 54.41 for 'Standard', 53.95 for 'reward_low_FP_rate', and 69.61 for 'reward_low_FN_rate', indicates improvements of 3.67%, 4.45%, and 2.63%, respectively, compared to the average scores of the NAB algorithms. The findings indicate that the proposed approach not only detects anomalies with high precision but also dynamically adapts to changing data characteristics without requiring manual recalibration. This study proposes a robust traffic anomaly detection method that ensures reliable monitoring and potentially facilitates effective traffic management and planning.The results of the study can be extended to other areas requiring real-time data monitoring and anomaly detection, offering a scalable solution adaptable to different contexts and requirements. © 2013 IEEE.