Enhanced Ensemble Machine Learning Technique to detect Bipolar Disorder

Abstract

The study introduced a unique Enhanced Ensemble Machine Learning (EEML) method for the detection of bipolar disorder, a serious medical diagnoses problem that acquires an immediate and timely diagnosis for an effective and earlier treatment. The EEML approach combines several machine learning models with the data based on patient’s attitude on different perspectives such as real, signal based, textual and behavioural data and more. In order to train individual classifiers for identifying mental diseases, pertinent characteristics are taken from each data source using a thorough feature selection technique. Relevance Vector Machine, Adaboost, Multi-Layer Perceptron and Recurrent Neural Network are among the classifiers; Random forest optimizes the classifiers. By capturing a greater variety of traits associated with mental diseases, the ensemble of classifiers improves overall performance. Three bipolar disorder datasets of various are used in the study to assess the EEML method on Dataset1 (a multimodal dataset), Dataset2 (signals on sensor-based dataset), and Dataset3 (real-time dataset). The EEML model excels with its higher accuracy of about 96.28% with Dataset1, 95.71% with Dataset2, and 98.5% with Dataset3. This study enhances the field of mental health diagnostics by taking advantage of ensemble machine learning's strengths to enhance detection accuracy and reliability.