Study of MLP-based classification of multi-cluster TPC signals

This study focuses on the classification of multi-cluster events based on a parameterization of data from a time projection chamber using machine learning. Samples containing a mixture of single and overlapping two-cluster events, both in one and two dimensions, were studied using multi-layer perceptrons and other MVA algorithms provided in the Scikit-learn package. The classification was based on various sets of features and classification accuracies of up to 97% for 1D clusters and 97% for 2D clusters were obtained. This study demonstrates that the efficient classification of signals for further processing through machine learning is feasible and efficient.