Floods have become more frequent and intense in recent years, resulting in an increasing toll on human lives and severe economic and environmental damage. In Rwanda, during the rainy season, floods disrupt several catchments, threaten lives, destroy infrastructure, erode farmland, and harm crops. According to the local community, there is no effective way to lower the risks of this disaster in their region. The goal of this study is to analyze weather data and model flood prediction using machine learning (ML) algorithms and find the best model that can be used to predict the flood sometime before its occurrence so that appropriate measures can be taken. It also investigated the source direction of rainfall that influences floods. Meteorological data for 38 years, flood cases, and water level variation in the Muvumba River were gathered from the relevant governmental institutions. The prediction of floods was conducted using three ML models, namely logistic regression (LR), random forest (RF), and Naïve Bayes (NB). Using Python packages, the correlation between rainfall from different stations and the river water level was identified. The results showed that the minimum temperature on the day of rainfall varied between 12.5 and 15 °C, whereas the maximum temperature was between 21 and 24 °C. The optimum value for humidity on the day of rainfall was 75–85%, while the atmospheric pressure was between 782 and 785 hPa. This correlation between daily weather data and daily rainfall indicates that these parameters can be used for short-term forecasting. The study also indicated the areas influencing a notable rise in water. These locations serve as significant sources of floodwater, providing valuable information for short-term flood forecasting purposes and flood prediction processes to warn the community. Based on the ML models’ performance, the NB model exhibited the highest true positive rate (TPR) at 91.4% and the lowest false negative rate (FNR) at 4.6%, whereas the TPR for LR and RF were 89.6 and 86.2%, respectively, and their respective FNR were 10.3 and 13.8%. This study’s findings are the benchmark for designing and developing an early warning system for flood prediction using the Internet of Things and an ML-based system that will provide real-time information to help the community, as well as decision-makers, make informed decisions on flood prevention and mitigation.