Engagement-Based Influencer Marketing Effectiveness: A Multi-Tier Analysis

MSc Computer Science Dissertation
Trinity College Dublin | 2025-2026

Overview

This repository contains the complete code, data processing pipelines, and analysis scripts for my MSc dissertation investigating influencer marketing effectiveness across different scales using machine learning.

Key Findings:

• Mid-tier YouTube influencers (100K–1M) achieved 3.29× higher engagement than mega-tier Instagram influencers (50M–500M)
• Machine learning achieved R²=0.39 for mid-tier prediction but failed (R²=-0.26) for mega-tier
• First systematic examination of scale-dependent predictability in influencer engagement

Research Questions

RQ1: What is the magnitude of engagement rate differences between mid-tier and mega-tier influencers?

RQ2: Which quantitative metrics best predict engagement at different scales?

RQ3: Does ML prediction accuracy vary systematically by influencer scale?

RQ4: What are the practical implications for influencer selection strategies?

##  Quick Start

### Prerequisites

- Python 3.10 or higher
- pip or conda package manager
- (Optional) YouTube Data API v3 key for data collection

### Installation

```bash
# Clone the repository
git clone https://github.com/YOUR_USERNAME/dissertation-influencer-ml.git
cd dissertation-influencer-ml

# Create virtual environment
python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

Running the Analysis

# Step 1: Data collection (requires API key)
python src/data/youtube_collector.py

# Step 2: Feature engineering
python src/data/feature_engineer.py

# Step 3: Run baseline models
python src/models/baseline.py

# Step 4: Train ML models
python src/models/ml_models.py

# Or run all notebooks in order (recommended for exploration)
jupyter notebook

Dependencies

Core Libraries

• Data Processing: pandas, numpy
• Machine Learning: scikit-learn, xgboost
• Visualization: matplotlib, seaborn, shap
• Statistical Analysis: scipy, statsmodels
• Data Collection: google-api-python-client

See requirements.txt for complete list with versions.

Key Results

Engagement Rate Comparison

Platform	Tier	n	Mean Engagement	Std Dev
YouTube	Mid (100K–1M)	45	3.65%	1.70%
Instagram	Mega (50M–500M)	40	1.11%	0.60%
Ratio	-	-	3.29×	-

Statistical Test: t(83)=6.75, p<0.001, Cohen's d=1.47

Machine Learning Performance

Model	YouTube R²	Instagram R²	Gap
Linear Regression	0.08	-0.15	0.23
Random Forest (Conservative)	0.39	-0.26	0.65
XGBoost	0.32	-0.79	1.11

Top Predictive Features (YouTube)

1. Comment-to-view ratio (r=0.382)
2. Video count (r=0.347)
3. Recent activity rate (r=0.297)
4. Total views (r=0.274)
5. Subscribers per video (r=0.268)

Methodology

Data Collection

• YouTube: YouTube Data API v3 (December 2024 – January 2025)
• Instagram: Kaggle "Top 200 Instagram Influencers" dataset
• Sample: n=85 (45 YouTube, 40 Instagram)

Feature Engineering

• Initial features: 21 (YouTube), 17 (Instagram)
• Final features: 8 (YouTube), 9 (Instagram)
• Systematic leakage detection and removal
• Samples-per-feature ratios: 5.6 (YouTube), 4.4 (Instagram)

Models Evaluated

• Baselines: Mean predictor, single-feature regression, full linear regression
• ML Models: Linear Regression, Ridge Regression (α=1.0, 5.0, 10.0), Random Forest (2 configs), XGBoost
• Validation: 5-fold cross-validation

Evaluation Metrics

• R² (coefficient of determination)
• RMSE (root mean squared error)
• MAE (mean absolute error)
• Train-test performance gap

Academic Context

Institution: Trinity College Dublin
School: Computer Science and Statistics
Degree: MSc in Computer Science
Course Code: CS7CS6 - Dissertation
Supervisor: Dr. Van-Dinh Nguyen
Submission: August 2026

Citation

If you use this code or reference this work, please cite:

@mastersthesis{varma2026influencer,
  author = {Varma, Shaurya},
  title = {Engagement-Based Influencer Marketing Effectiveness: A Multi-Tier Analysis of YouTube and Instagram Using Machine Learning},
  school = {Trinity College Dublin},
  year = {2026},
  type = {MSc Dissertation},
  note = {Available at: https://github.com/YOUR_USERNAME/dissertation-influencer-ml}
}

Data Access

Note: Raw data is not included in this repository due to:

• YouTube API Terms of Service (channel-level data)
• Kaggle dataset licensing (Instagram data)
• Privacy considerations

To replicate:

1. Obtain YouTube Data API v3 key from Google Cloud Console
2. Download Instagram dataset from Kaggle
3. Place in data/raw/ directory
4. Run data collection scripts

Processed features (with anonymized IDs) are available in data/processed/.

Acknowledgments

• Supervisor: Dr. Van-Dinh Nguyen for guidance and feedback
• Trinity College Dublin for research resources
• YouTube Data API v3 for data access
• Kaggle community for Instagram dataset

Contact

Shaurya Varma
MSc Computer Science Student
Trinity College Dublin
Email: [varmasl@tcd.ie]
GitHub: shauvar

---

Last Updated: April 2026