Machine learning model evaluation for 360° video caching

360° virtual reality videos enhance the viewing experience by giving a more immersive and interactive environment compared to traditional videos. These videos require large bandwidth to transmit. Typically, viewers observe only a part of the entire 360°videos, called the field of view (FoV), when watching 360°videos. Edge caching can be a good solution to optimize bandwidth utilization as well as improve user quality of experience (QoE). In this research, three machine learning models utilizing random forest, linear regression, and Bayesian regression have been proposed to develop a 360°-video caching algorithm. Tile frequency, user's view prediction probability and tile resolution have been used as feature. The purpose of the developed machine learning models is to determine the caching strategy of 360-degree video tiles. The models are capable to predict the viewing frequency of 360° video tiles (subsets of a full video). We have compared the results of the three developed models and the results show that the random forest regression model outperforms the other proposed models with a predictive R2value of 0.79.