Autism, a neurodevelopmental disorder, often presents challenges in emotional communication and regulation, impacting social interactions and overall well-being. Autistic children often express their emotions in unique ways compared to their non-autistic peers and may experience meltdowns more frequently as a response to overwhelming stimulations or emotional overload. Therefore, early detection and intervention are crucial to improving their well-being, especially in nurseries where many children are present. This paper proposes a reliable approach based on Deep Learning (DL) techniques to address potential emotional breakdowns in children with Autism Spectrum Disorder (ASD). It also addresses the pressing need for innovative solutions to support individuals with ASD in expressing and managing their emotions effectively. The ultimate goal is to forecast and avoid meltdown crises by creating a dependable and objective method for identifying early indicators of emotional distress in children diagnosed with ASD. The proposed model utilizes DL and transfer learning techniques to detect several emotions such as joy, sadness, anger, surprise, and natural emotion. When sadness or anger is detected, an alert message is sent to nursery staff who can benefit from the system as a tool for monitoring and supporting the children’s emotional development. The proposed model achieved an accuracy is 78.4% proving its ability to recognize the children’s emotions accurately based on available data.