Real-time discriminant analysis in the presence of label and measurement noise

Quadratic discriminant analysis (QDA) is a widely used classification technique. Based on training dataset, each class in the data characterized by an estimate of its center and shape, which can then be to assign unseen observations one classes. The traditional QDA rule relies empirical mean covariance matrix. Unfortunately, these estimators are sensitive label measurement noise often impairs model's predictive ability. Robust location scatter resistant this type contamination. However, they have prohibitive computational cost for large scale industrial experiments. We present novel method based recent real-time robust algorithm. additionally integrate anomaly detection step classify most atypical into separate outliers. Finally, we introduce bias plot, graphical display identify data. performance proposed approach illustrated simulation study with huge datasets, real datasets about diabetes fruit.