# Fusion of Transformer and RBF for Anomalous Traffic Detection in Sensor Networks

## TL;DR

This paper introduces RESTADM, a new model combining Transformer and RBF for detecting anomalies in sensor network data, achieving high accuracy on benchmark datasets.

## Contribution

RESTADM is a novel fusion model of Transformer and RBF for sensor anomaly detection, outperforming existing methods.

## Key findings

- RESTADM achieved an F1-score of 98.56% on the SMD dataset.
- The model scored 97.70% on the PSM dataset, surpassing traditional and modern baselines.
- The Transformer-RBF fusion proved effective for accurate and robust anomaly detection.

## Abstract

With the widespread adoption of the Internet of Things (IoT) and smart devices, the volume of data generated in sensor networks has increased dramatically, with diverse and structurally complex types that pose growing security risks. Anomaly detection in sensor networks has become a key technology for ensuring system stability and secure operation. This paper proposes a sensor anomaly detection model, termed RESTADM, which integrates a Transformer and a Radial Basis Function (RBF) neural network. The model first employs the Transformer to effectively capture the temporal dependencies in sensor data and then uses the RBF neural network to accurately identify anomalies. Experimental results on two public benchmark datasets, SMD and PSM, demonstrate the state-of-the-art performance of RESTADM. Our model achieves impressive F1-scores of 98.56% on SMD and 97.70% on PSM. This represents a statistically significant improvement compared to a range of baseline algorithms, including traditional models like CNN and LSTM, as well as the standard Transformer model. This validates the effectiveness of our proposed Transformer-RBF fusion, confirming the model’s high accuracy and robustness and offering an efficient security solution for intelligent sensing systems.