改进的Boruta算法在音乐情感研究中的应用

Application of improved Boruta algorithm in music emotion research

Boruta算法是一种新的特征选择算法,被设计成一个随机森林分类算法的包装器,包装算法的不同将导致不同的效果。对LightGBM算法进行包装,针对土耳其音乐数据集,使用随机森林、XGBoost以及LightGBM算法进行分类预测,通过计算准确率、Kappa系数及海明距离来判断各算法在该数据集上的适用性,结果显示,其性能优于使用随机森林和XGBoost算法生成的包装器。

Boruta algorithm is a new feature selection algorithm, which is designed as a wrapper of random forest classification algorithm. Different packaging algorithms will lead to different effects. The LightGBM algorithm is packaged. For the Türkiye music dataset, random forest, XGBoost and LightGBM algorithms are used for classification prediction. The applicability of each algorithm on the dataset is judged by calculating the accuracy, Kappa coefficient and Hamming distance. The results show that its performance is better than that of the wrapper generated using random forest and XGBoost algorithms.