viralscope

Term project for CS513 - Knowledge Discovery and Data Mining (Fall 2025)

GitHub Link: engindenizdogu/viralscope: Viral video predictor for Youtube videos

viralscope is a machine learning project that predicts YouTube video success using metadata and channel characteristics. The project processes 85+ million video records from the YouNiverse dataset to identify patterns that distinguish successful videos from others.

Key Features:

• End-to-end ML pipeline from raw data to trained models
• Random and data preparation sampling with engagement filtering
• Feature engineering with train/test split and scaling
• Multiple classification models with hyperparameter tuning
• Designed to prevent data leakage - labels created only after train/test split

Models with GridSearchCV:

1. Random Forest Classifier
2. Decision Tree Classifier
3. Linear SVC
4. K-Nearest Neighbors
5. Multi-Layer Perceptron