Optimization of Machine Learning for the Leptoquark Search using CERN ATLAS. Data

The Leptoquark is among the undiscovered particles which are being searched for in the Large Hadron Collider. Monte. Carlo simulated events of proton-to-proton collisions corresponding to the Leptoquark are studied with the ATLAS detector. The luminosity of the produced samples corresponds to the recorded data of 140 fb$^{−1}$. Four machine learning algorithms are used (TabNet, XGBoost, MLP, and Bayesian MLP) to train models to separate events on the 2lSS+1$\tau$ channel belonging to the. pair-production mode of Leptoquark from various background processes, including ttH, ttW, ttZ, tt, VV and other minor processes. The feature importance of the top performing models is constructed and utilized to produce more efficient models with improved sensitivity. In addition, the expected upper limit of cross-section for the pair-production of Leptoquark at 95% confidence level is calculated and compared to existing results.