Performance Analysis of FTRL algorithm for Dense and Deep Model of Music Mood Classification
Swati Sharma, Nipun Sharma, Anupama Sindgi, Anu Malhan, T. Prabhu, E. Suganya · 2024
The music mood classification problem is finding a lot of utility in the development of recommender systems designed related to health, exercising, listening and many others. Music induces change in the emotion or mood of a person. Analysis of effect of music on human electroencephalogram EEG reveals the stimulus induced. However, classification of the mood of the music is the preliminary step towards analyzing the effect of music. The preferred choice for such problems in Machine learning algorithm is KNN (K -Nearest Neighbors) Music mood classification is generally solved using the ADAM optimizer, and experimentation using FTRL is not done extensively. However, due to the fast convergence of FTRL, it can be tested for dense and deep networks too. This paper explores the performance of FTRL algorithm in music mood classification problem for calculating accuracy and loss. The results obtained in the paper are encouraging and matches the efficiency of KNN the performance of other optimizers like ADAM with an additional advantage of faster convergence.