Classification and detection of malware attack using hyperparameter optimization framework in SIoT

Abstract

Due to rapid advancement in World Wide Web and increased usage of internet, there is an extensive growth in Internet of Things (IoT). The IoT makes possible the operations smooth by enabling constant connectivity, granting internet access to all computing devices. Integrating and managing numerous devices is challenging with traditional methods. Nevertheless, social networks make it easier to establish communication between people. The Social Internet of Things (SIoT) enables objects to establish social connections based on human preferences. Diverse IoT devices interact to establish relationships by considering shared type of devices, attributes, and features. As Social Networks (SN) hold more and more data, they are more vulnerable to attacks by malware. As a result, malware detection is becoming a critical concern in the Internet of Things. Security can be significantly impacted by sociality in a number of ways. Advanced malware attacks must be identified using a method that is quick, dependable and efficient. This paper involves machine learning-based malware detection which looks threat identification at cyber security level and protection in the SIoT. The most advanced hyperparameter optimization framework (OPTUNA), which is used to detect malware from the Message Queuing Telemetry Transport (MQTT) dataset, was employed in this study. Furthermore, to regulate independent OPTUNA-based sampling algorithms, the proposed paper employs the Tree-Structured Parzen Estimator (TPE), which may be used to find quantization settings that maximize accuracy while minimizing latency. Moreover, the proposed OPTUNA based TPE sampler for LGBM model has accomplished the best Micro F1 score(accuracy) as 0.83 than the other classifier for identifying the types of malware attacks in SIoT.