Diagnostic Modelling for Bipolar Disorder Using Kinetic Activity logs using Machine Learning

The advancement carried out towards confirming the presence of Bipolar Disorder (BPD) among an individual is still evolving in slower pace in area of psychiatry. Review of existing literature show dominant adoption of varied machine learning approaches, which still has much wider scope of improvement. Hence, this manuscript presents a simplified computational model towards diagnosis of BPD considering kinetic activity logs of an individual. The initial step is towards improving the data quality where overfitting problems are mitigated using validation and distinct oversampling. The cleaned data is then transformed and subjected to compress the data without losing any significant information to generate potential feature. The obtained features when subjected to proposed neural network based machine learning generates optimal classification performance in contrast to conventional learning techniques.