• 主题
高级搜索  >
历史搜索 清空历史
首页> 外文期刊>Journal of Intelligent Information Systems >Granular methods in automatic music genre classification: a case study
【24h】

Granular methods in automatic music genre classification: a case study

机译：音乐类型自动分类中的细化方法：案例研究

获取原文
获取原文并翻译 | 示例
           
页面导航
  • 摘要
  • 著录项
  • 相似文献
  • 相关主题

摘要

Classification of music files by using the characteristics of the songs based on its genre is a very popular application of machine learning. The focus of this work is on automatic music genre classification based on granular computing methods (fuzzy rough, rough and near sets). We have proposed a modified form of supervised learning algorithm based on tolerance near sets (TCL 2.0) with a goal of exploring the scalability of the learning algorithm to a well researched music database composed of several genres. In the tolerance near set method, tolerance classes are directly induced from the dataset using the tolerance level epsilon and a distance function. We have compared the tolerance-based near set algorithm to a family of nearest neighbour (NN) algorithms based on fuzzy rough methods (FRNN) available in the WEKA platform. In terms of performance, the classification accuracy of TCL 2.0 is identical to the Bayesian Networks (BN) Algorithm, and comparable with the Sequential Minimal Optimization (SMO) Algorithm. However, the average classification accuracy of FRNN algorithms and the classical rough sets algorithm is better than TCL 2.0, BN and SMO algorithms. For this dataset, any accuracy over 90% is considered a very good classification accuracy which is achieved by all tested classifiers in this work.
机译：通过基于歌曲流派的歌曲特征对音乐文件进行分类是机器学习的一种非常流行的应用。这项工作的重点是基于粒度计算方法（模糊粗糙集，粗糙集和近集）的自动音乐流派分类。我们提出了一种基于容忍近集（TCL 2.0）的监督学习算法的改进形式，目的是探索学习算法对由多种流派组成的经过深入研究的音乐数据库的可扩展性。在公差接近设置方法中，使用公差等级epsilon和距离函数直接从数据集中导出公差等级。我们已经将基于容差的近集算法与基于WEKA平台中可用的模糊粗糙方法（FRNN）的最近邻（NN）系列算法进行了比较。在性能方面，TCL 2.0的分类准确性与贝叶斯网络（BN）算法相同，并且与顺序最小优化（SMO）算法相当。但是，FRNN算法和经典粗糙集算法的平均分类精度优于TCL 2.0，BN和SMO算法。对于此数据集，任何超过90％的准确度都被认为是非常好的分类准确度，这项工作中所有测试过的分类器都可以达到该准确度。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱：kefu@zhangqiaokeyan.com

京公网安备：11010802029741号 ICP备案号：京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有

  • 客服微信

  • 服务号