Towards a More Competent Healthcare System: A Methodological Approach for Quantizing Reliable Detection of Concealed Depression using Ensemble Learning

Depression is the most common psychological disorder, which effects a large number of people across the world. Depression needs to diagnosed earlier to make sure that a person receives timely counselling otherwise it can lead to severe problems. Moreover, stress and depression are not considered as a mental ailment instigating a substantial depressed population, unaddressed and untreated. Depression may further responsible for developing various chronic diseases and raises suicidal urges. This study has examined machine learning classifiers to categorize the depressed person from other, based on noted socio-demographic and psychosocial evidences. The dataset's most pertinent attributes were mined using ML algorithms. To improve the efficacy in prediction, Cross fold validation has been employed to lessen the training data's class imbalance. The ensemble technique has outperformed with an accuracy of 87 % & 88 % with or without using SMOTE respectively. Besides, the following evaluation measures, namely sensitivity, specificity, precision, F1-score, accuracy have been calculated for six different machine learning models to identify the most efficient model for predicting depression. This research also proposes a methodology that integrates federated learning with machine learning to monitor mental health utilizing distributed training of data. The global model resides on the global server. Local devices of users inherit global model to test their daily used keyboard data. After testing, the user's test data is anonymously sent to the global dictionary and subsequently updates the global dictionary using all user anonymous tested data. Then, using this revised global dictionary global model is trained once more and sent to all user local devices to track mental health. The use of Federated learning shows a visible improvement in accuracy with 89 % while keeping the client's data private.