My current project is an exercise implementing a binary classifier. It is trained to distinguish between two types of particle collision events in particle physics. These events are abundant in B meson experiments such as BELLE. The result can be used to compute the strength of these interactions. Current theoretical and experimental values do not agree, and higher precision measurement would allow a tighter constraint.
The existing algorithm employed by experiments is based on boosted decision trees (BDT). The new classifier is a 5 fully connected layers Keras neural network (NN) trained with simulated Monte Carlo (MC) data. Testing is done on data from a different MC generator for closure. Preliminary result shows that the NN outperforms the benchmark BDT but I am struggling to choose the best classification threshold for further investigation.