
Why White-Box Models in Enterprise Data Science Work More Efficiently
Ryohei Fujimaki, PhD | dotData CEO – discusses five key factors why white-box data science models are superior to black-box models for deriving business value from data science.
Data science will help organizations transform from data to knowledge, driving performance and competitive advantage. Data science platforms and methodologies should use the white-box model approach or the black-box model approach. The industry standard for machine-learning projects, black-box testing, is often a lack of actionable insights, leading to a lack of accountability. In this article, dotData founder, CEO, and Ph.D. Ryohei Fujimaki discusses the key factors why white-box models are superior to black-box models for data science projects.
There are two types of machine learning models: linear and nonlinear models. Non-linear models (black-box models) are opaque while white-box models are transparent. Data scientists create complex features and black boxes; they create nonlinear transformations, and deep learning (neural networks) computationally generates features; this makes the model a black box. ML model consumers need to understand why and how models predict to optimize their operations. White-box models typically give a reason for their predictions, whereas black-box models may only give a probability.
Enterprise data-science projects require that models are explainable; White box models provide explanations alongside prediction results. In data science, white-box models help organizations stay accountable for their data-driven decisions, whereas black-box models make it more difficult to remain compliant with the law.
Read the full article on e-Week: Why White-Box Models in Enterprise Data Science Work More Efficiently