YouTube Video Performance Prediction System Using Machine Learning and YouTube Data API

This paper presents an intelligent system designed to predict the performance of YouTube videos after publication. Most creators do not have a good way to determine how well their uploaded videos will perform because most of the tools available provide you with statistics of past performance but do not provide you with statistics of future performance that are reasonably accurate in order to assist these creators with estimating how their video or videos will do. The way this problem will be solved is by using the YouTube Data API to collect real-time data from YouTube in conjunction with using historical performance data from already uploaded videos so that the system will use machine learning to predict key performance indicators (KPI) such as views, likes, comments, virality score, and audience retention in the first seven days after being posted. The algorithms which will be used for predicting performance will be Random Forest and XGBoost. Also, trend analysis will be used to increase the accuracy of the data collected. Other techniques such as data pre-processing and feature engineering will be applied to improve the quality of the predicted performance. An interface has been developed using Flask to enable user interaction and visual representation of results from predicted performances. The performance of the algorithm proposed will be assessed based on the following performance indicators: MAE, RMSE, MAPE and R². Results obtained from the experimental evaluation have shown that the hybrid ensemble model provides more stable performance and produces more accurate predictions as compared to any of the single machine learning models used in the hybrid ensemble algorithm. This proposed framework will be useful for digital marketing and for content creators on YouTube.