Autism Spectrum Disorder (ASD) is marked by difficulties in communication and social interaction. Several studies have emphasized the importance of specific diagnoses to support professionals in the field, both in clinical diagnosis and therapies. This study investigates whether reducing the number of electrodes in low-density EEG systems, combined with machine learning techniques, can enhance the diagnosis of ASD. Using the publicly available dataset from the University of Sheffield, containing EEG signals from individuals with ASD across various age groups, our approach combines a proprietary interpolation method to estimate missing data with manual selection of 9 electrodes, reducing the initial quantity of 64 channels from the original dataset. Among the various tree and decision forest models trained, our best classifier achieved an accuracy of 98%. By contributing to the diagnosis of ASD more efficiently and affordably, this research aims to improve early intervention and potentially enhance outcomes and quality of life.