FIPS Moments for "A machine learning approach to classifying MESSENGER FIPS proton spectra"

Kappa distribution fits to FIPS proton spectra for the paper titled "A machine learning approach to classifying MESSENGER FIPS proton spectra" in the Journal of Geophysical Research: Space Physics.<br>This dataset contains the three parameters fitted to each proton spectrum (density, temperature and kappa). Each spectral fit is analysed using 9 artificial neural networks (ANNs), where ANNs 0-7 each analyse an eight of a spectrum and the final network (ANN 8) provides an overall assessment of the quality of fit. Each network outputs a probability of the fit being good and a corresponding class label. A spectral fit is determined to be "good" when the probability ≥ 0.5. The accuracy of the final neural network is 96%, so in some cases a few spectra may be misclassified - manual verification is recommended when using these data. Please read the aforementioned paper for more information.<br>The file "FIPSProtonClass.dat" is organised into 22 columns:<br>Column 1: Date and time of spectrum formatted as "yyyy-mm-ddTHH:MM:SS.sss", where yyyy is the year, mm is the month, dd is the day, HH is the hour, MM is minutes and SS.sss is seconds.Column 2: Proton density, <i>n_k, </i>in cm^-3.Column 3: Proton temperature, <i>T_k</i>, in MK.Column 4: Kappa value - note that any value over ~10 may be approximated using a Maxwellian distribution, and any value less than 1.5 is below the limit of anti-equilibrium and is not physical.Columns 5-12: Probabilities, P0-P7, that the kappa distribution fit to each spectrum is good (output from ANN 0-7 in the paper).Column 13: Probability that the overall fit is good (output from ANN 8 in the paper).Column 14-21: Class of each split section, where 0 is bad and 1 is good. The class is 1 when P ≥ 0.5.Column 22: Overall spectral class predicted by ANN 8, where 1 is good and 0 is bad.<br>