Scholarlink

<p>The proposed paper presents a new fuzzy ensemble learning method, FVELM, which is the first fuzzy ensemble learning approach to introduce a probabilistic S-shaped fuzzy activation function into an ensemble of Extreme Learning Machines (ELMs) for diagnosing autism spectrum disorder (ASD). We target two particular problems of neuroimaging analysis: the variability of models with randomly chosen parameters and the disruptive effect of outliers in fMRI data. The suggested FVELM model is the only one to integrate the deep feature extraction through the transfer learning method on a pre-trained ResNet-50 architecture with a strong, non-iterative fuzzy classifier. This study employs transfer learning and ensembles of Extreme Learning Machines (ELMs) to develop a new activation function for detecting autism using fMRI data. Comprehensive validation on the Autism Brain Imaging Data Exchange (ABIDE) I dataset shows that our framework is a new state-of-the-art, with a classification accuracy of 77.10 percent (95% CI: 74.5%–79.7%) and a 77.07% F1-score on a dataset of 1,035 individuals, 1.90 percentage points better than the previous best on the same sample size and dataset (ABIDE-I, n = 1035). This paper creates a new, powerful paradigm of ML-helping in multifaceted mental health diagnostics.</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

Integration of transfer learning and ensembles of extreme learning machines with fuzzy activation function for autism diagnosis

  • 期刊论文

  • 发表日期：

    2026年5月12日

查看全文
  • Daquan Chang

摘要

The proposed paper presents a new fuzzy ensemble learning method, FVELM, which is the first fuzzy ensemble learning approach to introduce a probabilistic S-shaped fuzzy activation function into an ensemble of Extreme Learning Machines (ELMs) for diagnosing autism spectrum disorder (ASD). We target two particular problems of neuroimaging analysis: the variability of models with randomly chosen parameters and the disruptive effect of outliers in fMRI data. The suggested FVELM model is the only one to integrate the deep feature extraction through the transfer learning method on a pre-trained ResNet-50 architecture with a strong, non-iterative fuzzy classifier. This study employs transfer learning and ensembles of Extreme Learning Machines (ELMs) to develop a new activation function for detecting autism using fMRI data. Comprehensive validation on the Autism Brain Imaging Data Exchange (ABIDE) I dataset shows that our framework is a new state-of-the-art, with a classification accuracy of 77.10 percent (95% CI: 74.5%–79.7%) and a 77.07% F1-score on a dataset of 1,035 individuals, 1.90 percentage points better than the previous best on the same sample size and dataset (ABIDE-I, n = 1035). This paper creates a new, powerful paradigm of ML-helping in multifaceted mental health diagnostics.