Predicting video conversion time from video metadata and conversion parameters using gradient boosting machine

Predictive analytics techniques can tremendously improve the performance of computing systems by optimizing energy, waiting time and throughput via predicting the execution time of scheduled jobs beforehand. As a consequence of the correlation between video conversion parameters and video conversion time, the conversion time is highly predictable from input video properties and conversion parameters. Hence, this paper proposes gradient boosting machine to predict the conversion time of videos using video metadata and conversion features with no detailed information about the applied codec. The evaluation results of the experiments conducted on benchmark Youtube video characteristics dataset showed that our model reduced the conversion time prediction error by as much as 4.03% over previously applied models. The proposed model also indicates that features about coding standard and codec allocated memory used for conversion, size, duration, bitrate and framerate of videos are crucial for conversion time prediction.