Application of Machine Learning for the Higgs Boson Mass Reconstruction Using ATLAS Data

This thesis deals with the reconstruction of the Higgs boson mass decaying in the 2lSS + 1$\tau_{\rm had}$ channel in the ttH production. Based on the reconstructed mass, the goal is to separate the signal from background productions such as the ttZ. The data created by the full ATLAS detector simulation are used to develop two neural networks. First, a classification neural network that organizes the data by assigning detected particles to corresponding positions in the channel. Second, a regression neural network that reconstructs the mass of the Higgs boson. The developed neural network is then tested on different data selections and is shown to outperform the Missing Mass Calculator technique. Finally, the neural network is tested on real ATLAS data.