[Paper Review] Applying Machine Learning in Self-Adaptive Systems: A Systematic Literature Review

Omid Gheibi, Danny Weyns+1 more|arXiv (Cornell University)|Mar 6, 2021

Advanced Software Engineering Methodologies54 references41 citations

TL;DR

A systematic literature review of 109 studies on applying ML to architecture-based self-adaptive systems with MAPE loops, detailing motivations, methods, and open challenges.

ABSTRACT

Recently, we witness a rapid increase in the use of machine learning in self-adaptive systems. Machine learning has been used for a variety of reasons, ranging from learning a model of the environment of a system during operation to filtering large sets of possible configurations before analysing them. While a body of work on the use of machine learning in self-adaptive systems exists, there is currently no systematic overview of this area. Such overview is important for researchers to understand the state of the art and direct future research efforts. This paper reports the results of a systematic literature review that aims at providing such an overview. We focus on self-adaptive systems that are based on a traditional Monitor-Analyze-Plan-Execute feedback loop (MAPE). The research questions are centred on the problems that motivate the use of machine learning in self-adaptive systems, the key engineering aspects of learning in self-adaptation, and open challenges. The search resulted in 6709 papers, of which 109 were retained for data collection. Analysis of the collected data shows that machine learning is mostly used for updating adaptation rules and policies to improve system qualities, and managing resources to better balance qualities and resources. These problems are primarily solved using supervised and interactive learning with classification, regression and reinforcement learning as the dominant methods. Surprisingly, unsupervised learning that naturally fits automation is only applied in a small number of studies. Key open challenges in this area include the performance of learning, managing the effects of learning, and dealing with more complex types of goals. From the insights derived from this systematic literature review we outline an initial design process for applying machine learning in self-adaptive systems that are based on MAPE feedback loops.

Motivation & Objective

• Understand the problems that motivate the use of machine learning in self-adaptive systems with MAPE-based feedback loops.
• Identify key engineering aspects of integrating learning with self-adaptation (MAPE functions, learning dimensions, methods).
• Characterize learning problems and their role within adaptation problems.
• Highlight open challenges and propose an initial design process for applying ML in MAPE-based SAS.

Proposed method

• Systematic literature review following a predefined protocol (planning, execution, reporting).
• Automatic search across IEEE Xplore, ACM DL, and Springer Link, complemented by manual refinement using a pilot query.
• Inclusion criteria: 2003–May 2020 publications, ML applied to MAPE-based self-adaptive systems, with basic evaluation.
• Exclusion criteria: surveys, tutorials, short papers, editorials.
• Data extraction with 14 items (authors, year, title, venue, citation count, quality score, adaptation problem, learning problem, MAPE functions, learning dimensions, learning methods, domain, limitations, challenges).
• Analysis via descriptive statistics and open coding for qualitative categories.

Experimental results

Research questions

• RQ1 RQ1: What problems have been tackled by machine learning in self-adaptive systems?
• RQ2 RQ2: What are the key engineering aspects considered when applying learning in self-adaptation?
• RQ3 RQ3: What are open challenges for using machine learning in self-adaptive systems?

Key findings

• 6709 papers were identified initially, with 109 retained for data collection.
• Selected papers span 75 venues.
• Most studies were published from 2015–2019 (72%), with 28% between 2007–2014.
• Reporting quality across studies was generally sufficient for problem description and context, with structured evaluation.
• ML is primarily used to update adaptation rules/policies and to manage resources for better quality-resource balance.
• Supervised and interactive learning with classification, regression, and reinforcement learning are dominant; unsupervised learning is comparatively rare.
• Learning is commonly used to assist the analyzer (e.g., reducing adaptation options) and to support runtime decision-making within MAPE loops.
• Open challenges include learning performance, handling learning-induced effects, and dealing with more complex goals.