<p>Autistic spectrum disorder (ASD) is a neurological condition characterized by difficulties in social interaction, communication, and repetitive behaviors. Despite its largely hereditary nature, early detection is crucial, and a potential strategy for more efficient and quicker diagnosis is to employ artificial intelligence (AI). In this paper, a new system is introduced that is called automatic screening autism (ASA) system. ASA comprises three primary stages: data preparation (DP), feature selection (FS), and patient detection (PD). In the first stage, the used dataset is preprocessed through several steps: handling missing values and rejecting outliers. Then, these preprocessed features are fed to the FS stage to select the most important features using improved genetic algorithm (IGA). IGA composed of two stages; (i) pre-selection stage (PS²) using information gain (IG) and (ii) conclusive selection stage (CS²) using genetic algorithm (GA). Subsequently, these attributes are input into the proposed classification model using optimized deep neural network (ODNN). Actually, ODNN is based on optimized weights of traditional DNN using various optimization algorithms, and the final decision is derived from the best performance. ASA has been evaluated against contemporary methodologies. Results from experimental studies demonstrate that the proposed ASA outperforms its competitors regarding accuracy, precision, sensitivity, and F-measure, achieving values of approximately 99.10, 98.90, 98.70, and 98.80% in that order.</p>