Smile prediction has enormous promising demand for tracking mental health providing information about a person’s emotional wellness. This study suggests a unique method for predicting smiles in mental health monitoring by tracking smile expression and correlating it with emotional states using video sequences. This work has created a more reliable and accurate model for evaluating mental health by combining facial recognition for smile prediction. The proposed research work uses deep CNN along with the ResNet-50 architecture for smile prediction using CK+ data set. The study explores the potential uses of smile prediction systems in clinical contexts such as the early identification of anxiety, depression and emotional discomfort. The deep CNN and ResNet-50 architectural model solves the issues of overfitting and vanishing gradient. The proposed work is tested on CK+ datasets and contrasted with current techniques, the suggested system demonstrates that this proposed approach performs better than other conventional models and attained the results of 95%.