Enhanced Anomaly Detection in Automotive Systems Using SAAD: Statistical Aggregated Anomaly Detection

TL;DR

The paper addresses the challenge of accurate, robust anomaly detection in automotive sensor data by proposing Statistical Aggregated Anomaly Detection (SAAD), which fuses a histogram-based statistical labeling with a dropout-augmented Fully Connected Network (FCN). The approach uses artificial labeling and a learning-based predictor, then combines their outputs via a rules-based aggregation framework that leverages the DL confidence $v$ and thresholds $a$ and $b$ to produce final decisions. Empirical results on Hardware-in-the-Loop automotive data show standalone accuracies of 72.1% (statistical) and 71.5% (DL), with aggregation improving to 88.3% accuracy and an F1 score of 0.921, demonstrating the method’s effectiveness and robustness. The findings suggest SAAD’s potential for real-time, high-stakes anomaly detection in automotive systems and beyond, with future work focusing on adaptive thresholds and multivariate time-series extensions.

Abstract

This paper presents a novel anomaly detection methodology termed Statistical Aggregated Anomaly Detection (SAAD). The SAAD approach integrates advanced statistical techniques with machine learning, and its efficacy is demonstrated through validation on real sensor data from a Hardware-in-the-Loop (HIL) environment within the automotive domain. The key innovation of SAAD lies in its ability to significantly enhance the accuracy and robustness of anomaly detection when combined with Fully Connected Networks (FCNs) augmented by dropout layers. Comprehensive experimental evaluations indicate that the standalone statistical method achieves an accuracy of 72.1%, whereas the deep learning model alone attains an accuracy of 71.5%. In contrast, the aggregated method achieves a superior accuracy of 88.3% and an F1 score of 0.921, thereby outperforming the individual models. These results underscore the effectiveness of SAAD, demonstrating its potential for broad application in various domains, including automotive systems.

Figures (4)

• Figure 1: Architecture of the SAAD(Statistical Aggregated Anomaly Detection) framework, showcasing the integration of statistical anomaly detection with a Fully Connected Network (FCN) for enhanced anomaly detection in automotive systems. The architecture includes data preprocessing, artificial labeling, deep learning model training, and result aggregation to achieve improved accuracy and robustness. The effectiveness of SAAD is demonstrated with two samples, X and Y. The FCN initially predicts that both are anomalies, with sample X having high confidence and sample Y having low confidence. The statistical method, using $k=2$, detects sample X as an anomaly and sample Y as not. Aggregation is applied to obtain the final answer, depending on the outputs of the models and the confidence of the deep learning one, combining the strengths of both methods.
• Figure 2: Anomalous bins exemplification on histogram
• Figure 3: Heatmap of Accuracy Scores Obtained on Aggregated Results. The heatmap visualizes the accuracy percentages across different thresholds $a$ and $b$. The color intensity represents the accuracy, with warmer colors indicating higher accuracy scores. The pattern shows that accuracy generally increases as both thresholds $a$ and $b$ increase, with the highest accuracy observed in the lower right corner of the heatmap, where both thresholds are at their maximum values. This indicates a positive correlation between higher threshold values and improved accuracy.
• Figure 4: Evolution of accuracy at different threshold values.