Song Ranking Prediction Based On Random Fores

The emergence of digital technology has promoted the vigorous development of the music industry,which has brought about certain changes in the attribute characteristics,communication laws and consumption patterns of music itself.For the domestic traditional music industry,the new trend of transition from traditional digital song downloads to streaming song services also means that the era of transformation of China’s music industry has arrived.This dissertation mainly studies whether the genre and popularity of songs are related to the digital music attribute characteristics of songs in the current digital era,which has great commercial significance for the current music market and can provide certain reference value for music practitioners.This dissertation selects the US Billboard Hot 100 chart as the research object to analyze the digital music characteristics of the songs on the list and whether the related external dependent variables can help predict the song genre category and song at the highest ranking level of the chart,and determine whether the random forest algorithm is the best algorithm for predicting the highest ranking grade of the song.This dissertation first preprocessed the collected data and finally randomly selected 12,000 songs released between 1960 and 2019 as the dataset for research purposes.The focus of this dissertation is divided into two parts: one is to use SVM algorithm and other classification algorithms such as logistic regression and decision tree to classify the genres of music songs,and compare and evaluate the classification effects of each classification algorithm.The results show that the classification accuracy of SVM algorithm in the test set is the highest among the classification algorithms,and the F1 score is 0.7315.Therefore,determining SVM algorithm as the optimal algorithm for predicting the classification of music genres can help the music industry better predict whether a song belongs to the mainstream genre and propose meaningful estimates for the commercial value of songs.The second is to determine the song genre,set the target variable genre in one as an independent variable,and then add external dependent variables that affect the ranking of the song list,that is,the average ranking and the length of time on the list.Combining the above variables and 15 numeric feature variables,algorithms such as random forest and decision tree are used to classify and predict the highest ranking level of songs.The results show that the accuracy of the random forest algorithm on the test set reaches 74.465%,which is higher than the classification effect of the decision tree and other algorithms,and the classification accuracy of the random forest algorithm after feature selection is increased to 75.681%.Therefore,the random forest algorithm is selected as the optimal algorithm for predicting the highest ranking level of songs on the chart.In summary,the dissertation finds that the popularity of songs is related to acoustic characteristics,and the genre classification of songs is also related to acoustic characteristics.That is,in the era of streaming media,the digital music characteristics of digitized songs have substantial predictive value.When capital needs to choose a new song among the many newly released songs for investment,publicity and promotion,it can predict the highest ranking level of these songs in the future through the random forest algorithm after determining the genre to which the new song belongs,so as to help it choose the right song for investment and achieve a higher rate of return on commercial value.