Abstract
With the rapid development of information and communications technology, massive data are collected every day from every corner of a digitalized smart city. Mining hidden patterns from these data, therefore, becomes an important task for better monitoring, understanding, and management of a smart city. To achieve this, machine learning is a powerful tool. Unlike typical machine learning applications, such as image classification and text sentiment classification, pattern mining tasks in the smart city scenario often require extra domain knowledge and more case-specific design of how machine learning algorithms are applied. In this thesis, a general framework is first summarized for applying machine learning algorithms to extract hidden patterns from data collected in smart city applications. Afterward, the framework is applied to three real-world applications to better understand human behaviors in three important aspects of a livable city, namely transportation management, social event organization, and energy conservation. To tackle the challenges in each specific application, the best type of machine learning algorithm is selected, and the way of applying machine learning is carefully designed concerning case-specific requirements. First, classification algorithms are exploited to identify the transportation modes from large-volume, sparse, and unlabeled spatiotemporal data, which provides a precise yet low-cost solution to obtaining daily transportation information of large population in a smart city. Next, clustering algorithms are adopted to analyze data collected by WiFi-based passive sensor networks to understand crowd behaviors in a large outdoor social event, both spatially and temporally, which provides insights to the event organizer for better future management. Lastly, regression and clustering algorithms are integrated with domain knowledge to design an effective approach aimed at benchmarking the air-conditioning energy performance of residential rooms, which helps users to monitor and conserve the energy consumed by air-conditioning.