Signal extraction in atmospheric shower arrays designed for $\rm 200\,GeV-50\,TeV$ $\gamma$-ray astronomy.

We present the SEMLA (Signal Extraction using Machine Learning for ALTO) analysis method, developed for the detection of $\rm E>200\,GeV$ $\gamma$ rays in the context of the ALTO wide-field-of-view atmospheric shower array R&D project. The scientific focus of ALTO is extragalactic $\gamma$-ray astronomy, so primarily the detection of soft-spectrum $\gamma$-ray sources such as Active Galactic Nuclei and Gamma Ray Bursts. The current phase of the ALTO R&D project is the optimization of sensitivity for such sources and includes a number of ideas which are tested and evaluated through the analysis of dedicated Monte Carlo simulations and hardware testing. In this context, it is important to clarify how data are analysed and how results are being obtained. SEMLA takes advantage of machine learning and comprises four stages: initial event cleaning (stage A), filtering out of poorly reconstructed $\gamma$-ray events (stage B), followed by $\gamma$-ray signal extraction from proton background events (stage C) and finally reconstructing the energy of the events (stage D). The performance achieved through SEMLA is evaluated in terms of the angular, shower core position, and energy resolution, together with the effective detection area, and background suppression. Our methodology can be easily generalized to any experiment, provided that the signal extraction variables for the specific analysis project are considered.