Machine Learning Approaches to Improve Security and Performance Monitoring of IoT Devices

Kayode, Olumide
Journal Title
Journal ISSN
Volume Title

In this current era of Internet of Things (IoT), data privacy and security of Internet enabled devices has become a major concern of many users and device manufacturers. Massive amount of data is being generated by these IoT devices and there might be possibilities of user's information being exposed without any privacy protection. The rate of data transfer, size, kind of information transmitted and secure channels used by these IoT devices are of utmost importance and demand more exploratory research. Moreover, the "always on" and "always connected" attributes of IoT devices necessitate working condition as well as performance monitoring. Unexpected downtime and sudden breakdown of IoT devices can be extremely destructive especially for safety-critical systems. Condition monitoring and health state estimation are vital techniques for maintaining high reliability. Effective approach to investigate security and privacy of wide range of IoT devices needs to be developed. Using a proxy server, we investigate the data being transmitted by six representative IoT devices, analyze the data and propose an intelligent approach for proxy connection monitoring. Our results show that user's information and devices' identities were being leaked in our experiments. The applied neural network classifier uses network connection information to effectively detect proxy connections and performs better than Support Vector Machine as well as logistic regression models that were developed. We further propose a robust proxy detection mechanism suit-able for stochastic and deterministic malicious alteration of connection information. The approach is based on Deep Q-Network and Generative Adversarial Network. For condition monitoring, we propose a lightweight model operable on edge device for Remaining Useful Life (RUL) estimation. The model aptly utilizes the time series sensor data and successfully predicts the remaining useful life. Towards a distributed estimator in smart home environment, we also developed a model based on Long Short Term Memory (LSTM) neural network for estimating energy utilization. These research works demonstrate excellent results and contribution to knowledge. Our work addressed two major challenges in IoT, namely security and performance monitoring. The various data driven approaches and methods that we developed can be applied to enhance data security and performance monitoring in IoT. Security mechanisms to detect unsolicited proxy connection, anomalies or cyber attacks have been proposed. Furthermore, our techniques for estimating remaining useful life and energy utilization in smart home environment are effective. Efficient method for distributed learning and use case are also proposed to illustrate its feasibility. These are approaches that can improve reliability, performance monitoring and time-critical data driven computation.

This item is available only to currently enrolled UTSA students, faculty or staff.
Data security, Deep reinforcement learning, Internet of Things, Machine learning, Performance monitoring, Smart home
Computer Science