Proteus: A scalable, flexible and extensible multi-classifier framework

Abstract

Though the popularity and demand for machine learning infrastructures is soaring in this age of “big data”, general purpose configuration and deployment strategies are still in their infancy. This paper presents Proteus, a flexible and extensible framework allowing different machine learning algorithms to be introduced in a plug-and-play manner in order to be evaluated. Proteus enables domain experts to more easily compare, contrast, and even combine results from classifiers including Deep Learning, GLM, GBM, Naive Bayes, Random Forest, SVM and Linear Regression. Leveraging this design, it is easier to explore the possibility that a combination of multiple classifiers may be the best approach to guaranteeing high accuracy. A case study involving 6 months of mouse-movement data from 5 patients with a Clinical Dementia Rating (CDR) of 0 (control group) and 5 patients with a CDR of 0.5 (considered a high impairment level) identifies the costs and benefits of this engineering effort towards a scalable, flexible and extensible architecture for multi-classifier analysis.