<p>The immersive experience of watching movies presents a unique opportunity to observe and analyze adults’ facial expressions and emotional responses in a naturalistic environment. The significant growth of the entertainment industry in recent years has expanded the availability and personalization of movies and other media, offering a platform for studying behavioral patterns in diverse populations. This study explores the feasibility of diagnosing Autism Spectrum Disorder (ASD) in adults through the analysis of facial images captured during movie watching sessions. A Kaggle dataset comprises 2653 facial images of individuals diagnosed with ASD and neurotypical adults, focusing on emotion-related expression changes over time to develop a diagnostic model. To address this challenge, we propose the integration of deep learning methodologies to classify ASD in adults aged 18 to 30 years. The data set was pre-processed to standardize and enhance the variability of the training images, including resizing to a fixed target size, horizontal flipping for augmentation and rescaling to normalize pixel values. These pre-processed images are input for a hybrid model that combines multi-convolutional neural networks (multi-CNN) and bidirectional long- and short-term memory (BiLSTM) networks. The multi-CNN component efficiently extracts spatial features from the images, while the BiLSTM leverages temporal dynamics, capturing the sequential nature of changes in facial expressions over time. The proposed model achieved an accuracy of 96.6% in predicting ASD diagnoses, demonstrating the efficacy of the work done. This high level of performance underscores the potential of integrating computer vision and deep learning techniques for non-invasive, scalable diagnostic tools. The findings contribute to the development of efficient methods for the detection of ASD in young adults, emphasizing the importance of emotion-related behavioral analysis in autism research.</p>