Early detection of disease using electronic health records and fisher's wishart discriminant analysis

Linear Discriminant Analysis (LDA) is a simple and effective technique for pattern classification, while it is also widely-used for early detection of diseases using Electronic Health Records (EHR) data. However, the performance of LDA for EHR data classification is frequently affected by two main factors: ill-posed estimation of LDA parameters (e.g., covariance matrix), and “linear inseparability” of the EHR data for classification. To handle these two issues, in this paper, we propose a novel classifier FWDA -- Fisher's Wishart Discriminant Analysis, which is developed as a faster and robust nonlinear classifier. Specifically, FWDA first surrogates the distribution of “potential” inverse covariance matrix estimates using a Wishart distribution estimated from the training data. Then, FWDA samples a group of inverse covariance matrices from the Wishart distribution, predicts using LDA classifiers based on the sampled inverse covariance matrices, and “weighted-averages” the prediction results via Bayesian Voting scheme. The weights for voting are optimally updated to adapt each new input data, so as to enable the nonlinear classification.