Social media has become the primary form of communication wherein users can share intimate moments online through photos, videos, or posts. At a glance, while this greatly improves interconnectivity between people, it also increases the propensity towards unrestricted acts of Cyberbullying, prompting the need for a data-centric detection system. Unfortunately, these sites generate much metadata, which begs the need for complex Machine Learning (ML) classifiers to categorize these acts accurately. Prior studies on the subject matter only target the topics of Conceptualization, Characterization, and Classification of Cyberbullying individually, so this research aims to provide a more holistic understanding of the subject matter in a continuous, synthesized format. This study found that Cyberbullying differs from Traditional Bullying in key areas of Repetition and Intention. Moreover, multimodal feature sets, as opposed to single feature sets, significantly improve ML classifiers' performance. Lastly, the selection of appropriate ML classifiers and performance metrics is context-dependent. The result of this study presents a consolidated view of relevant parties tackling different aspects of an ML-based automated Cyberbullying detection system so that those assigned tasks can approach them strategically