Enhancing Smart Cities Through Real-Time Insights and Safety: A Comparative Study of Supervised Machine Learning Algorithms for Anomaly Detection in Emerging Urban Landscapes

Abstract

Today, IoT devices contribute to the creation of intelligent environments, encompassing structures like smart buildings, hospitals, banks, houses, and offices. With the increasing prevalence of smart environments, the importance of data from IoT devices becomes evident. Collecting IoT data within smart cities presents a more intricate challenge. Data acquired through sensors and devices often falls prey to corruption or damage, resulting in anomalies or anomalous data. These anomalies have a substantial influence on the functionality of smart cities. Enhancing infrastructure autonomy, safety, and efficiency is indispensable for elevating citizens' quality of life. Swift identification and correction of anomalies play an important role in achieving this goal. Anomaly detection assumes a critical role in the seamless operation of smart cities, driven by multiple factors. Firstly, it offers early detection of potential issues, empowering city officials to intervene before problems exacerbate, thereby averting harm or disruption. Secondly, proactive maintenance can address impending concerns, curtailing their escalation into major disruptions, thus bolstering infrastructure reliability and reducing downtime. Thirdly, optimizing resource consumption, including energy usage, leads to superior resource management and waste reduction. Fourthly, anomaly detection is paramount for spotting security threats, enabling swift countermeasures to safeguard citizens and vital infrastructure. Lastly, anomaly detection offers invaluable insights into smart city operations, facilitating data-driven decisions that enhance efficiency, safety, and citizen well-being. The core objective of an anomaly exposure system revolves around differentiating between authentic and deceptive activities. Machine learning techniques have emerged as a common approach for identifying data anomalies, enhancing performance across diverse systems. This research specifically targets anomaly identification within smart cities. Initial data collection involves sourcing information from varied outlets to encompass a broad spectrum of smart city datasets. Subsequently, a suite of machine learning algorithms is tested to assess their performance. Python serves as the platform for applying these algorithms, which include Naïve Bayes, K-Nearest Neighbor, Gradient Boosting, Decision Tree, and Random Forest. The results garnered from machine learning algorithms prove to be both precise and efficient. The study's performance assessment metrics encompass accuracy, recall, precision, and F-measure. By comprehensively analyzing machine learning techniques across datasets, a comparative assessment is achieved. Remarkably, different techniques display varying levels of proficiency across individual datasets. The findings indicate that Naïve Bayes and Decision Tree achieved the highest average accuracy of 91.75% across all datasets. In terms of recall, Random Forest demonstrated an average of 93% across datasets. For precision and F-measure, Naïve Bayes excelled with scores of 99.1% and 0.95 respectively.