Post-Partum Depression (PPD) is a significant medical condition that occurs in some women after childbirth as a consequence of physiological behavioral and mental alterations. The complicated symptoms of this condition make it difficult to diagnose and differentiate from other conditions. Timely detection and diagnosis of PPD are crucial for controlling morality rates and ensuring effective treatment. Various Machine Learning (ML) models was developed to predict PPD based on the patients demographic status, mental health history and vital signs. But, additional psychological attributes are needed for predicting mental status and identifying individuals with PPD risk. Also, costeffective and innovative methods are needed to identify individuals with PPD and detect potential development tendencies. Hence, in this paper, Osprey Parameter Optimized MLP (OPOMLP) is developed to address the above issues for efficient PPD. Initially, the Application Programming Interface (API) function of online social network like Twitter (tweets) and Instagram (comments) are used to collect the data posted by health care professionals. Then, Natural Language Processing (NLP) is utilized to pre-process the collected data \and extract relevant text of Twitter users to estimate PPD phases. The additional psychological attributes like mental health and behavioural changes attributes of women are extracted using Linguistic Inquiry Word Count (LIWC) and Latent Semantic Analysis (LSA) methods. Next, MLP network is trained using the extracted attributes along with the attributes of demographic status, mental health history and vital signs. Since, the parameters of MLP were not optimized properly which leads to computational complexities in the PPD prediction. So, the weights initialization and the hyper-parameters of MLP is optimized simultaneously by using an Osprey Optimization Algorithm (OOA). OOA is a metaheuristic optimization algorithm derived from osprey bird hunting behavior which aims to find the global optimum solutions and reduces the complex optimization issues. The relationship between hyperparameters and classification performance will identify an optimal hyperparameter space regions for optimal classification with less computational time and resources. Finally, the OPOMLP is employed for the final prediction of PPD. The test outcomes reveal that the OPOMLP model achieves an accuracy of 96.12% on the collected dataset compared to the classical PPD detection models.