Neural Network-based Prediction of Solar Activities

A data mining system designed to associate previous solar flares with sunspot groups using databases from Nobeyama Radioheliograph and the National Geophysical Data Center, and associated data used for training verification, and comparison of several Neural Networks topologies which can be used with an automated solar activity prediction system in the future. The data mining system manages to associate 272 flare and sunspot groups out of 1628 flares and 65363 sunspot groups using the degree of correspondence between their locations and time data. The cascade feed forward backpropagation trained network provided the optimum performance and 85% correct prediction for the possible occurrence of a solar flare was obtained. In addition, 78% of the class of the occurring flares are predicted correctly.