This study investigates the prediction of lightning flash rates using a ground-based, S-band polarimetric weather radar and storm cellbased lightning prediction. Both short-term cases during multiple years and long-term cases are used for training and testing by combining the radar and ground-based lightning detection network measurements. The identified lightning-generating storm cells are tracked during the data pre-processing to initiate our machine learning (ML) algorithm. The selection of features for ML algorithm implementation is discussed, and the performance of lightning rate prediction is evaluated using Pearson's correlation coefficient (r) and mean percentage error (MPE) metrics. The results show that using polarimetric measurement can improve prediction accuracy with short-term data sets, while for long-term datasets, the improvement is less significant. In general, reasonable prediction accuracy is achieved for up to 30 minutes of lead time.