Machine Learning Application for Particle Identification in MPD

work presents results of the first tests of machine learning application using gradient boosting on oblivious decision trees to particle identification problem in Multi Purpose Detector (MPD) experiment on Nuclotron based Ion Collider fAcility (NICA) at Joint Institute for Nuclear Research. Categorical boosting (CatBoost) implementation of a gradient boosting on decision trees has been used. Particle identification was based on the information provided by the time projection chamber (TPC) and the time-of -flight (TOF) subdetectors. In the study three various Monte-Carlo datasets of measurements from TPC and TOF were simulated and used within CatBoost classifiers training and testing. The comparison was made with the n-sigma method which is currently used at MPD software. Gradient boosting shows better efficiency in case of small and large momentum values (p < 0.7 GeV/c and p > 1.5 GeV/c). This demonstrated that machine learning methods are well suited to address the particle identification problem in MPD experiment.