Machine Learning with System/Software Engineering in Selection and Integration of Intelligent Algorithms

Abstract

Machine learning has become an essential component in the design of intelligent systems across several disciplines. This widespread use of machine learning has led to the importance of evaluating how Systems/Software Engineering approaches go hand in hand with Machine Learning to reliably integrate intelligence in software systems. In this research effort, our motivation is to develop a systematic approach also termed as Machine Learning Engineering for the selection and integration of machine learning algorithms in systems. The proposed approach discusses combining a structured approach for designing and developing system/software with an experimental analysis that data scientists perform on machine learning algorithms. This experimental analysis is essential as some of the characteristics exhibited by intelligent algorithms cannot be predicted or guaranteed compared to systems without intelligent algorithms. In this paper, we elaborate on our system/software engineering guided disciplined approach by comparing two machine learning algorithms that focus on the recognition of handwritten digits. The algorithms we compare are the Logistic Regression and Neural Network algorithms. After the analysis, we identify the contracts that should be associated with intelligent components to better predict the behavior of the system as a result of the selection of one of the components to be a machine learning algorithm. Finally, we indicate how the results can be used by Systems/Software Engineers in integrating intelligent algorithms.