User Centric Social Opinion and Clinical Behavioural Model for Depression Detection

In more recent time, depression as a lingering mental illness as continued to affect the way people act, and behave consciously or otherwise. Though it remained an undiagnosed disease globally without prejudice to age, gender, color or race; a lot of people never know implicitly or explicitly when they are depressed until it begins to affect their health conditions. While depression can be deciphered through text analysis in opinion mining, oftentimes, changes in human body also provides a convincing status of a depressed individual. No doubt, each data source can independently predict human depression status; however, the exclusive mutual relationship between both data sources has not been studied for depression detection. Therefore, in identifying meaningful correlations between clinical and behavioural data, this research detected depression by analyzing and matching mined patterns in users’ behavioural opinion through tweets with trackable changes in clinical body vitals using wearable device for effective therapy in depressed patient management. Thus, by using a 5-fold cross validation on the clustered data, Random Forest ensemble model was used to build the Social-Health Depression Detection Model (SH2DM) after data preprocessing and optimal feature extraction. The dual data sourced user-centric model produced a better predictive result in accuracy, precision and recall values when compared and evaluated with single data depression detection instances of clinical and behavioural records.