Multivariate methods for particle identification

The purpose of this project was to evaluate several MultiVariate methods in order to determine which one, if any, offers better results in Particle Identification (PID) than a simple n$\sigma$ cut on the response of the ALICE PID detectors. The particles considered in the analysis were Pions, Kaons and Protons and the detectors used were TPC and TOF. When used with the same input n$\sigma$ variables, the results show similar perfoance between the Rectangular Cuts Optimization method and the simple n$\sigma$ cuts. The method MLP and BDT show poor results for certain ranges of momentum. The KNN method is the best performing, showing similar results for Pions and Protons as the Cuts method, and better results for Kaons. The extension of the methods to include additional input variables leads to poor results, related to instabilities still to be investigated.