The Identification of Conflicting Determinations of Anomalies in Computer Network Behaviour: Cyber Security in Smart Cities

New problems and difficulties have arisen as a result of smart cities' greater reliance on various forms of urban technology. Data-driven technologies and computer networks are used in a broad range of applications and municipal infrastructures. Intrusion detection systems are always looking for evidence of malicious activity on computer networks. Binary classifiers are good tools for determining whether data should be considered normal or pathological. The removal of redundant and unimportant data may be accomplished via the use of feature selection and feature scaling. Nine of the twenty-four numerical characteristics included within the Kyoto 2006+ dataset are regarded as being crucial for the training of models. This linearization is accomplished via a hyperbolic tangent normalization. Different machine learning techniques with proven classification abilities were utilized in this study. While there was a little decline in accuracy overall (less than 0.1 percent), there was a significant reduction in the amount of time it took to analyze data. According to the findings, there is an undeniable advantage to using TH scaling in terms of processing time. It doesn't matter how accurate the classifiers are; there are still situations when their conclusions aren't the same. Based on the XOR process achieved on the results of two separate classifications, we develop a conflicting decision detector. These classifiers are a feedforward neural network, which is the quickest, and a weighted k-nearest neighbor model, which is the most accurate but also the slowest. The findings indicate that up to 6% of the various choices made may be identified.