Thunderstorms are well-known severe weather phenomena of the Gangetic West Bengal (GWB) region of India. The objective of the present study is to identify the ranges of Max_Z parameters of Doppler Weather Radar (DWR) associated with precipitating clouds that eventually grow into thunderstorms and to obtain a model to assess the predictability of thunderstorm and non-thunderstorm events with maximum possible accuracy during the pre-monsoon season (April–May) over the metropolis Kolkata (22.6°N; 88.4°E) enclosed within GWB (20–26°N, 85–91°E), India. The DWR imageries are analyzed to identify the stages of thunderstorm development. The survival of the fittest principle of genetic algorithm (GA) is implemented to find a suitable combination of the DWR Max_Z parameters; the reflectivity, distance of the first detected echo from Kolkata where the DWR is installed and the echo top height for the genesis of thunderstorms. The problem is posed as an optimization problem and the values of the parameters are converted into binary strings. The result reveals that the echoes with reflectivity between 44 and 48 dBZ at a distance of 250–300 km from Kolkata with echo top height between 13 and 15 km have the maximum possibility to grow into a thunderstorm. The artificial neural network (ANN) model is developed with the values of the Max_Z parameters optimized by GA as the inputs. The target of the ANN model is to forecast the type of the echo cells leading either to thunderstorm or non-thunderstorm events. The result further reveals that the ANN model with three hidden layers and one node in each layer is the most suitable model for estimating the likelihood of thunderstorm/non-thunderstorm events with mean absolute error (MAE) of 0.71/2.81. The result of the study is validated with the observation of India Meteorological Department. © 2014, Springer-Verlag Wien.