<p>Individuals with Autism Spectrum Disorder (ASD) often exhibit challenges in recognizing emotions from facial expressions, hindering social interactions. Emotion recognition systems have emerged as vital tools for improving human–computer interaction, particularly benefiting individuals with ASD who face challenges in interpreting nonverbal communication such as facial expressions and body language. This study explores the development and implementation of emotion recognition system which will be an aid to individuals with ASD to face realworld challenges. The work uses ResNet-50 deep learning algorithm combined with meta-learner and attention mechanisms for emotion recognition. The models are trained and tested on the facial expression recognition (FER) and real-world affective faces database (RAFDB) datasets. The ResNet-50 model with meta-learner achieved a training accuracy of 95.49% and a validation accuracy of 92.92%. The ResNet-50 model with attention mechanisms slightly outperformed the meta-learner model, achieving a training accuracy of 97.08% and a validation accuracy of 93.04%. To corroborate the performance of deep learning models, confusion metrics is evaluated from experimental results. The work carried out can facilitate better social interactions and improve the quality of life for individuals with ASD.</p>