Auto encoders with Cellular Automata for Anomaly Detection

Abstract

This work combines auto encoders with cellular automata (CA) to present a novel hybrid strategy for anomaly identification. For feature learning, auto encoders are used to identify spatial patterns in the input data. Simultaneously, temporal and geographical dependencies are captured by CA, which improves the model's capacity to identify complicated anomalies. Training on spatially altered data, the auto encoder-CA hybrid model makes use of CA's temporal evolution to reveal dynamic patterns. Reconstruction errors between the input data and its decoded representation are computed to identify anomalies. A comprehensive framework for anomaly identification is provided by the synergy between spatial-temporal analysis of CA and auto encoder-based feature learning. Performance is optimized by fine-tuning the model's parameters, which include the auto encoder architecture and CA setup. The hybrid model's dynamic adaptation ensures robustness over time by accommodating changing data distributions. Evaluation measures show how well the suggested method captures abnormalities that appear in both temporal and geographical dimensions. A promising method for identifying abnormalities in complicated datasets with detailed spatial and temporal patterns is presented by this novel combination of auto encoders and cellular automata. The proposed method is evaluated with various parameters like reconstruction error, precision , recall, F1 score and Area Under the Receiver Operating Characteristic (ROC-AUC). The average accuracy is reported as 97.36% which is promising when compared with baseline methods.