Automated Framework for Abnormal Activity Detection & Smart Security System

Abstract

As a human being, our brain is always tuned to spotting something out of the “normal” or the “usual stuff”, separate those anomalies which don't coordinate the ordinary layout. Information science strategies regularly look for deviations that don't buy in to the typical progression of information through the complete development of information. For example, a "abnormally large" number of login endeavors may prompt a potential digital assault, or in a concise measure of time, a huge spike in installment card buys might be Visa extortion. The IoT network is increasingly feared for IoT assaults and anomalies. Through increased use of IoT technologies in each environment, types of attacks in all of these technologies often increase proportionately. Service denial, data type checking, malicious control, malicious activity, filtering, surveillance and incorrect set-up are attacks that may cause an IoT device failure. Their use in many realistic applications, such as entertainment, healthcare, simulation and surveillance systems, has increased due to human fraud, assault and anomaly detection. Wearable devices have been used for typical applications used to monitor or recognize human behavior. Such applications, however, require the individual's physical touch.

Machine learning solution that can recognize and protect the device when it is abnormal is proposed to address these challenges. Several classifications have been used for this purpose and ML algorithms such as Logistic Regression (LR), Vector Machine Support (SVM), Decision Tree (DT), Random Forestry (RF) and Artificial Neural Network (ANN). Vision based attack and identification of anomalies are becoming more beneficial as no human interference or physical interaction with humans is required. In addition, a range of cameras are networked to monitor and recognize the agent’s activities. The measures used to assess the system performance are also accuracy, recall, f1 measure as well as the area below the recipient's operational similarity function. By using our proposed system, it gains more accuracy & Security without human interaction.