Rainfall variability affects agricultural production, water resources, and flood risk in regions vulnerable to extreme weather events. This study examines rainfall behavior and flood forecasting in the most flood-affected coastal area of Bangladesh in the Sylhet region using machine learning approaches which may provide inputs for early warning and disaster preparedness. The analysis uses two different datasets: a monthly rainfall dataset for predicting short-term fluctuations and a 44-year monsoon-specific database to quantify long-term rainfall changes. Using ensemble models such as Random Forest, Gradient Boosting, XGBoost, and Support Vector Regression, we were able to predict rainfall with high accuracy. The Random Forest model emerged as the most precise model, particularly in the months from June to October, when floods frequently occur across the nation. On the other hand, support vector regression has limitations because of its lower precision when dealing with irregular rainfall variations. The results revealed that the seasonal rainfall from June to October differed from that from November to February, exhibiting strong peak increases during the monsoon period, which was likely due to flood events and potential socio-economic disruptions in Sylhet. The small size of our training set stunted the performance of maximum fit, despite the striking trend we observed on the monsoon dataset, highlighting the importance of very long datasets for reliable predictive power. Research on rainfall prediction in the Sylhet region based periodically on historical weather data forms a basic framework of knowledge that will help with flood prediction and planning sustainable communities as climate change worsens. Variability affects agricultural production, water resources, and flood risk in regions vulnerable to extreme weather events.