Cyberbullying using digital platforms to harm others is a global issue with significant emotional and psychological impacts, often linked to suicide. While social media fosters global connectivity, it also enables harmful behaviors like cyberbullying. With Bangla being the seventh most spoken language globally, the growing number of Banglaspeaking online users calls for a dedicated detection system. This study presents a novel approach for detecting Bangla cyberbullying by utilizing a custom-built Own dataset of realworld Bangla social media comments, specifically collected for this research. Annotations include a dataset identifying harmful content. Psychological frameworks are applied to analyze the comments, and word embedding techniques are used for text representation. The classification is carried out using the Random Forest algorithm, achieving 95.78% accuracy in detecting cyberbullying. The Work also the impact of userspecific factors such as location, age, gender, and engagement on detection accuracy. The results suggest that incorporating these elements can enhance detection, making Random Forest a strong candidate for real-world cyberbullying detection in Bangla.