YouTube Trending Video Predictor

Predicting Virality Through Machine Learning

Prabesh Aryal · Rahma Ibrahim · Piyush Singh · Cristobal Talamantes · Gloria Abolade

---

What Are We Investigating?

• What factors drive YouTube videos to trend and achieve massive visibility on the platform?
• Can AI accurately predict a video's trending potential using only its metadata, before it's published?
• Can we build an A/B testing simulator to compare YouTube video titles and optimize content before posting?

Why YouTube Virality?

Billions of Daily Users — YouTube's trending page shapes what content reaches massive audiences, making it one of the most influential discovery engines online.

Real-World Impact — Creators, educators, journalists, and marketers all depend on platform reach. Understanding virality bridges data science with media and communication.

Video Is Dominant — Video is the dominant form of online communication. Predicting engagement has never been more relevant — or more valuable.

Core Research Question

"How much of a video's popularity can be explained by its metadata alone?"

We approach this question through two complementary framings:

Framing	Goal
Regression	Predict exact view counts — a continuous target variable using metadata features
Classification	Categorize videos as low / medium / high popularity — a discrete, interpretable output

Dataset

Data Sources

• YouTube Trending Video Statistics — Kaggle (datasnaek/youtube-new)
• YouTube Trending Videos – Daily Update — Kaggle (canerkonuk)

Global Coverage

US · UK · Germany · Canada · France · Russia · Mexico · South Korea · Japan · India

Key Features

• View count, likes, dislikes, comments
• Category, tags, channel title
• Publish time and engagement metrics

How Data Will Be Used

• Analyze YouTube trending data: views, engagement, categories, tags, publish time
• Train ML models to find key drivers of trending performance
• Help creators optimize content and publishing strategies
• Inform marketers, researchers, and recommendation systems
• Show how ML uncovers trends and supports data-driven decisions

Models

Baseline — Linear Regression

• Predicts view count from metadata (publish time, category, tags, likes, comments)
• Shows feature–popularity relationships
• Serves as an interpretable benchmark for advanced models

Advanced — Random Forest

• Identifies complex, nonlinear patterns between video metadata and YouTube popularity
• Captures interactions among variables like views, likes, comments, categories, tags, and publish time
• Feature importance scores reveal which factors most heavily influence trending performance

Recognizing & Addressing Bias

Sources of Bias

1. Sampling Bias — Datasets include only trending videos, over-representing successful content.
2. Geographic Bias — Some regions have more data; models may not generalize equally across countries.
3. Historical Bias — Past data may not reflect current platform algorithms or creator trends.

Mitigation Strategies

1. Showcase Results — Explain to the user why the model reached its conclusion.
2. Break It Down — Analyze results separately across categories, regions, and platforms.
3. Be Transparent — Clearly communicate that the model identifies patterns but cannot guarantee whether a post will go viral.

Project Structure

HitOrFlop/
├── data/                  # Raw YouTube trending video datasets
├── cleaning.py            # Data cleaning and preprocessing
├── eda.py                 # Exploratory data analysis
├── featureExtraction.py   # Feature engineering from cleaned data
└── README.md

Getting Started

1. Download the dataset from Kaggle and place it in the data/ directory.
2. Run the data pipeline in order:

python cleaning.py
python eda.py
python featureExtraction.py

References

1. Canerkonuk. (n.d.). YouTube Trending Videos Global [Data set]. Kaggle. https://www.kaggle.com/datasets/canerkonuk/youtube-trending-videos-global/data
2. Sprout Social. (n.d.). YouTube Trends. https://sproutsocial.com/insights/youtube-trends/
3. Chen, Y.-L., & Chang, C.-L. (2019). Early prediction of the future popularity of uploaded videos. Expert Systems with Applications, 133, 59–74. https://doi.org/10.1016/j.eswa.2019.05.015
4. Aljamea, A., & Zeng, X.-J. (2024). Predicting the Popularity of YouTube Videos: A Data-Driven Approach. In S. Prajapat, P. Grace, L. Yang, P. Jenkins, & N. Naik (Eds.), Advances in Computational Intelligence Systems (Vol. 1453, pp. 625–639). Springer International Publishing AG. https://doi.org/10.1007/978-3-031-47508-5_48
5. Stokowiec, W., Trzcinski, T., Wolk, K., Marasek, K., & Rokita, P. (2017). Shallow reading with Deep Learning: Predicting popularity of online content using only its title. arXiv. https://doi.org/10.48550/arxiv.1707.06806