The volume of social media data generated over the past few years became the source of a vast amount of user-generated data. It can be used to understand people's mental state from the extracted insights. The objective is to create an automatic, scalable approach to proficiently recognize disturbing patterns connected with psychological well-being through social media. This proposed approach focuses on the extraction of social media inputs indicating the first signs of mental disorders such as anxiety and depression. This model uses machine learning algorithms and preprocessing is carried out to make the data suitable for model building. Then emotional states are identified using nouns through sentiment analysis, emotion classification, and text mining. The input dataset is the collection of tweets used in labeling and we have used six emotions to label the tweets such as fear, anger, surprise, sad, happiness, and love. Leveraging the data-based method for preliminary detection and intervention, this research develops and interprets the TPOT (Tree-based Pipeline Optimization Tool) AutoML for digital mental wellness concerning social media data. This analysis revealed that, from all the applied algorithms, Logistic Regression resulted in the highest accuracy of 85%. The XGBoost method resulted in an accuracy of 82%.