人工蜂群算法优化支持向量机及其在音乐流派自动分类中的应用

Using ABC Algorithm Optimizing Support Vector Machines with Its Application in Musical Genre Automatic Classification

音乐流派是区分和描述不同音乐的一种标签,借助数学和计算机的方法将大量音乐自动分为不同流派是目前国内外研究的热点问题之一.支持向量机(SVM)由于其具有严格的数学理论基础而被广泛应用于音乐流派自动分类.然而,支持向量机的惩罚参数和核参数对其分类效果具有重要影响.以交叉验证正确率作为适应值,采用人工蜂群(ABC)算法优化支持向量的控制参数.在音乐流派自动分类的仿真实验中,经ABC算法优化后的支持向量机取得的平均预测正确率为80.8000%(最优预测正确率达83%),高出默认参数SVM 18.8个百分点.与粒子群优化算法及遗传算法相比,仿真实验结果同样显示了ABC算法的优越性.

Musical Genres are labels used to distinguish and describe different pieces of mu- sic. It is now one of the hot topics both at home and abroad that automatically classifying large amount of music according to their genres with the help of mathematics and comput- ers. Support vector machines（SVM） are widely used in musical genre automatic classification since it is strictly supported by mathematical theory. However, the penalty and nuclear pa- rameters in SVM have significant impact on its effects of classification. This paper uses cross validation accuracy as the fitness value, and adopts Artificial Bee Colony （ABC） algorithm to optimize control parameters of SVM. In the simulation experiment of musical genre auto- matic classification, the optimized SVM gets 80.8000% mean prediction accuracy（and the best prediction accuracy is up to 83%） which exceeds the SVM using default parameters by 18.8 percent. Compared to Particle Swarm Optimization （PSO） and Genetic Algorithm （GA）, the experimental results also show the superiority of ABC algorithm.