الفهرس 
Machine Learning AlgorithmsOnly 14 pages are availabe for public view
 |  | from | 161 |  |  |
| | | from | 161 | | |
Abstract
In recent years, Industrial Control System (ICS) includes several categories of control systems such as Programmable Logic Controllers (PLC), Supervisory Control and Data Acquisition (SCADA) systems and Distributed Control Systems (DCS). These control systems are found in the industrial sectors and critical infrastructures such as Gas Pipelines, transportation networks, electric power distribution networks, gas, water distribution networks and nuclear power generation. Therefore, they can be vulnerable to several types of internal and external cyber-attacks. The study in this work is provided with analysis of a system based on classification methods applied for anomaly detection over Industrial Internet of Things (IIoT) systems specifically in water SCADA systems. Machine learning methods are also used to investigate cyber-attacks found in SCADA systems. The classification and detection of intrusions and attacks will help system administrators and industrial operators to take the prime decision to protect critical industrial systems from hackers and malicious behaviors as well as provide detecting anomaly activities including physical component failures, and sabotage. Likewise, the proposed model is used to predict the cyber-attacks which may happen in the IIoT systems through raising an alarm in the presence of an attack against the industrial system and help the system administrators to detect digital evidence in an effective manner.
In this thesis, we introduce four proposed works as follows: The target of the first proposed work is to explore and study of Machine Learning algorithms for malicious activities recognition in critical industrial water infrastructure while the second proposed model is enhancement and resolving problems in the malicious activities detection system to increase the evaluation metrics to be suitable for real-time water-based systems. The evaluation metrics for K-Nearest Neighbors (KNN) classifier are the best values therefore, we propose the third model which is composed of different optimization techniques with KNN machine learning algorithm. The performance metrics of Optimized KNN model are the highest values compared to the state-of-the-art models and the other models proposed previously in this thesis. Finally, the fourth proposed model is composed of Optimized KNN model and Synthetic Minority Oversampling Technique (SMOTE) which enhanced the performance metrics such as accuracy, Precision, Recall and F1-score.