This study develops a machine learning framework to determine whether a mesoscale convective system (MCS) will evolve into a tropical cyclone (TC), aiming at deeper understanding the genesis mechanism of the TCs. It builds on an MCS database that is established by Huang et. al. using brightness temperature imagery from the Cloud Archive User Service (CLAUS) project to identify and track MCSs. The machine learning algorithms extract essential properties associated with MCSs, such as vertical wind shear, instability, sea surface temperature, profiles of humidity and vorticity, and predict the possibility of becoming TCs in a few days. The predictive capability is achieved by establishing meaningful spatial and temporal relationships among the properties along the track of individual MCSs through exhaustive machine learning. We also explore the lead time in predicting a TC genesis using this method. The preliminary results show that the machine learning method is able to capture the distinct features of the TC genesis. This framework can be applied to other weather and climate phenomena, such as classifying pre-stage conditions leading to MCS genesis and different categories of TCs/hurricanes.