Modeling Data Requirements for Machine Learning Systems

Abstract

As machine learning technology penetrates into various fields, how to ensure the quality of machine learning systems becomes an urgent problem. Current requirements modeling methods for machine learning systems are still in their infancy and rarely include data requirements modeling. In this paper, we propose a two-layer data requirements modeling method for machine learning systems. The bottom layer is the learning context used to describe the elements of the machine learning system and environment and relationships between them. A feature-oriented domain analysis approach is used to describe the learning context with feature models, and give the definitions, relationships and constraints of features. The upper layer is a set of property-based specifications. The definition of features and the descriptions of feature relationships provide the basis for the construction of properties. We derive a set of properties to be satisfied on the basis of the constructed learning context, and based on this we give descriptions and specifications of the data requirements for the machine learning systems. To better demonstrate the approach, we use an example of a self-driving system throughout the article.