Scholarlink

<p>Inspired by the concepts of granular computing and interval computing, this paper focuses on applying the concept of information granulation to the probabilistic neural network (PNN) to reduce its structural complexity and build an interval granular PNN (IGPNN) classifier with the ability to handle both numerical and interval data. The information granules occupying the certain areas in the feature space are generated by clustering the numerical training data through the fuzzy C-means (FCM) algorithm. Here we give explicit methods to quantify and improve the homogeneity level of the information granules, and to convert them into interval form. Then the interval information granules are used to construct the pattern layer of IGPNN, so that the built network has a simplified structure and the ability to handle both numerical and interval data. A series of simulations based on a synthetic dataset and nine public datasets demonstrate that IGPNN has the advantages of simple structure, low computational and storage resource utilization compared to traditional PNNs. In addition, a real-world application is presented to further validate the practical performance of the proposed classifier.</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

Interval granular probabilistic neural network for unified processing of classification data

  • 期刊论文

  • 发表日期：

    2026年1月20日

查看全文
  • Haohong Huang,
  • Shouping Guan,
  • Qiuyang Fang

摘要

Inspired by the concepts of granular computing and interval computing, this paper focuses on applying the concept of information granulation to the probabilistic neural network (PNN) to reduce its structural complexity and build an interval granular PNN (IGPNN) classifier with the ability to handle both numerical and interval data. The information granules occupying the certain areas in the feature space are generated by clustering the numerical training data through the fuzzy C-means (FCM) algorithm. Here we give explicit methods to quantify and improve the homogeneity level of the information granules, and to convert them into interval form. Then the interval information granules are used to construct the pattern layer of IGPNN, so that the built network has a simplified structure and the ability to handle both numerical and interval data. A series of simulations based on a synthetic dataset and nine public datasets demonstrate that IGPNN has the advantages of simple structure, low computational and storage resource utilization compared to traditional PNNs. In addition, a real-world application is presented to further validate the practical performance of the proposed classifier.