Constant-Time Linear Regression Learning and Its Applications on Real-Time R&M Systems

Constant-time linear regression algorithms are presented in this paper. Machine learning techniques rely on a good amount of data to achieve high performance. However, processing big data is time-consuming and thus limits the potential of applying machine learning techniques to real-time R&M (Reliability and Maintainability) systems. Linear regression has a long history in R&M applications, which also serves as a basic machine learning method. Most linear regression applications conduct the learnings off-line, not in real time, which causes the resulting model to be outdated. Realizing that continuous learning is crucial in real-time operations, this study is devoted to constant-time algorithms findings. Both lifelong and moving-time-window constant-time linear regression algorithms are presented. The case study demonstrates the benefits of applying the constant-time algorithms to real-time system monitoring.

Automated summary

This paper presents constant-time linear regression algorithms and discusses the potential and limitations of applying machine learning techniques to real-time system monitoring.