Scholarlink

<p>Commonly, the MAPE-K (Monitor, Analyzer, Planner, and Executor over Shared Knowledge) loop serves as the standard foundational framework for self-adaptive systems (SASs). The integration of Deep Learning (DL) into the Analyzer has significantly enhanced decision-making efficacy. However, creating DL models requires considerable expertise and time, often with extensive manual supervision. Moreover, in dynamic environments, a practical challenge emerges when models face unseen data, impeding real-time adaptability. This frequently necessitates rebuilding models from scratch to sustain Analyzer performance and reliability. In this paper, we explore the use of Automated Machine Learning (AutoML) to automate the development and selection of DL models for the DeltaIoT SAS dataset. Basically, we use AutoGluon, a state-of-the-art open-source framework, to classify our tabular dataset. Specifically, we focus on two DL architectures: NeuralNetFastAI (FastAI) and NeuralNetTorch (NN_Torch). To the best of our knowledge, this study is the first to apply AutoGluon to the DeltaIoT dataset. Experimental evaluations have shown that AutoGluon not only reduces manual intervention but also outperforms manually optimized models. The best AutoGluon-generated DL model attains an accuracy of up to 94%, an F1-score of 91%, an MCC of 87%, and a G-Mean close to 94%.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Optimized automated analysis using AutoGluon-driven deep learning for advancing self-adaptive systems

  • 期刊论文

  • 发表日期：

    2025年10月15日

查看全文
  • Sabah Lecheheb,
  • Soufiane Boulehouache,
  • Said Brahimi

摘要

Commonly, the MAPE-K (Monitor, Analyzer, Planner, and Executor over Shared Knowledge) loop serves as the standard foundational framework for self-adaptive systems (SASs). The integration of Deep Learning (DL) into the Analyzer has significantly enhanced decision-making efficacy. However, creating DL models requires considerable expertise and time, often with extensive manual supervision. Moreover, in dynamic environments, a practical challenge emerges when models face unseen data, impeding real-time adaptability. This frequently necessitates rebuilding models from scratch to sustain Analyzer performance and reliability. In this paper, we explore the use of Automated Machine Learning (AutoML) to automate the development and selection of DL models for the DeltaIoT SAS dataset. Basically, we use AutoGluon, a state-of-the-art open-source framework, to classify our tabular dataset. Specifically, we focus on two DL architectures: NeuralNetFastAI (FastAI) and NeuralNetTorch (NN_Torch). To the best of our knowledge, this study is the first to apply AutoGluon to the DeltaIoT dataset. Experimental evaluations have shown that AutoGluon not only reduces manual intervention but also outperforms manually optimized models. The best AutoGluon-generated DL model attains an accuracy of up to 94%, an F1-score of 91%, an MCC of 87%, and a G-Mean close to 94%.