abarela13/youtube_regression_project

Estimate viewership for newer channels

Project 2: Regression - YouTube Growth Predictor

Identifying what affects channel viewership for small channels

Abstract

The goal of this project was to use regression models to predict the total channel viewership of fairly "new" youtube channels (less than 2 year) in order to project possible revenue for channels with monetization. I worked with data scraped from ChannelCrawler and SocialBlade.

Data

The dataset ultimately contains 854 observations with 8 features for each.

• acquisition: web scraping
• storage: flat files
• sources:
  • ChannelCrawler
  • SocialBlade

Target variable:

• Total Channel Views

Features:

• Videos Published
• Channel Creation Date (Age)
• Channel Views
  • Total
  • Weekly
• Subscribers
• Total
• Weekly Gained/Lost
• Monetized

Algorithms:

• Simple Linear Regression
• Ridge Regression
• Lasso Regression

Tools:

• Web Scraping
  • BeautifulSoup, Selenium
• Data Cleanup
  • Pandas
• Modeling
  • scikit-learn, statsmodels

Analysis:

• Model: Linear Regression
• Feature engineering: StandardScaler,PolynomialFeatures
• Cross-Validation
• Regularization: Ridge, Lasso, RidgeCV, LassoCV