Deep learning approach to predict music track skips in the audio streaming service by car users

Abstract: With the advent of technology, the audio industry has transitioned to streaming services where users can listen to songs instead of having LPs or cassettes. Users of audio streaming services can access audio on a variety of devices, including Bluetooth speakers and headsets, TVs, gaming consoles, smartwatches, smartphones, automobiles, and more. Data collected in the process of giving streaming access to the user has the potential to improve the service over time. This thesis focuses on data from a single platform to predict a music track skips in audio streaming service by car users to see how much performance can be obtained. The thesis used Random Forest and Fully connected feedforward deep neural network (F-DNN) models. Various modifications to default hyperparameters are used to see if the performance of models can be increased. The outcome demonstrates that a model trained on data from a single platform can reasonably predict the skip of a music track.