In the wake of a historic political shift in Bangladesh—marked by the end of a 16-year authoritarian regime through mass student-led protests—social media platforms, particularly Facebook, have become a hotspot for politically charged discourse. This period has seen a sharp increase in hate speech, including aggressive posts, trending hashtags, and direct verbal attacks against political figures and parties. The widespread and unregulated nature of such content poses serious risks to social harmony, democratic engagement, and public trust. Consequently, there is a critical need for automated systems capable of detecting and categorizing political hate speech in Bangla. This study addresses this need by developing and evaluating a range of models—including traditional machine learning (ML), deep learning (DL), and advanced transformer-based architectures—to classify political hate speech into four categories: Pa-PoliHaS (Passive Political Hate Speech), Dir-PoliHaS (Direct Political Hate Speech), Non-PoliHaS (Non-Political Hate Speech), and Other. A novel dataset comprising 11,118 politically reactive Facebook comments was constructed for this purpose. Among the models tested, the Bangla BERT-SQuAD transformer model outperformed others, achieving the highest F1-score of 0.74. This research contributes toward building effective, scalable solutions for Bangla hate speech detection, particularly in politically sensitive and linguistically diverse digital environments.