A neural network was trained to accurately predict the entire single-event specific energy spectra for use in alpha-particle microdosimetry calculations. The network consisted of 4 inputs and 21 outputs and was trained on data calculated using Monte Carlo simulation where input parameters originated both from previously published data as well as randomly generated parameters that fell within a target range. The 4 inputs consisted of the source-target configuration (consisting of both cells in suspension and in tissue-like geometries), alpha particle energy (3.97-8.78 MeV), nuclei radius (2-10 μm), and cell radius (2.5-20 μm). The 21 output values consisted of the maximum specific energy (zmax), and 20 values of the single-event spectra, which were expressed as fractional values of zmax. The neural network consisted of two hidden layers with 10 and 26 nodes, respectively, with the loss function characterized as the mean square error (MSE) between the actual and predicted values for zmax and the spectral outputs. For the final network, the root mean square error (RMSE) values of zmax for training, validation and testing were 1.57 x10-2, 1.51 x 10-2 and 1.35 x 10-2, respectively. Similarly, the RMSE values of the spectral outputs were 0.201, 0.175 and 0.199, respectively. The correlation coefficient, R2, was > 0.98 between actual and predicted values from the neural network. In summary, the network was able to accurately reproduce alpha-particle single-event spectra for a wide range of source-target geometries.