With the rapid increase in the use of the Internet, cyber-attacks are escalating rapidly. One sophisticated cyber-attack is Advanced Persistent Threat (APT). An APT gains access to a system and remains there for a long time utilizing continuous, covert, and sophisticated hacking techniques. The targets of these attacks are often large corporations or government networks. The theft of confidential documents, stolen user credentials, and database loss are possible outcomes of this attack. The importance of APT detection has attracted a lot of research attention, so many methods have been proposed in the literature to address APT detection. However, most of the proposed methods have focused on only the first one or two phases of ac APT attack; one reason might be the lack of a comprehensive APT dataset covering all phases. Motivated by the use of honeypots to monitor hacking activities and machine learning (ML) capabilities to identify patterns from a dataset and make a logical decision, researchers have tried to generate logs and integrate ML to detect APTs. However, most people tried to implement a honeypot monitoring a single network service. The overall aim of this thesis is to improve the performance of APT detection by utilizing various ML techniques. Moreover, the focus of this thesis is to build ML models to predict APTs proactively and utilize versatile honeypots for various network services monitoring to make a comprehensive APT dataset. The preliminary work has been done, and the findings are promising compared to baseline and other ML methods.