Jira: AXMC-235

Objective

In the highly dynamic and competitive landscape of mobile advertising, understanding what factors drive user engagement is critical to optimizing media spend and maximizing return on investment. Our team conducted an in-depth analysis of mobile ad clicks using Principal Component Analysis (PCA), leveraging data stored in Amazon Redshift and RDS Aurora MySQL. The goal was to identify key attributes that influence ad clicks and impressions at scale.

Outcome

This PCA-driven approach to analyzing mobile ad clicks provided a scalable, data-driven methodology for optimizing digital advertising strategies. The integration of Redshift, RDS Aurora, and Tableau allowed for both large-scale data processing and intuitive visualization, ensuring that insights translated into actionable business decisions.

Data Collection & Storage

The source data consisted of terabytes of advertising transaction records, totaling hundreds of millions of rows. These records were stored in:

• Amazon Redshift for analytical querying of large datasets
• Amazon RDS Aurora MySQL for transactional integrity and real-time access

The dataset included a wide range of attributes capturing various aspects of ad media performance:

• Key Attributes Analyzed:
  • Geography (Country, City)
  • Ad Unit
  • Publisher/App Name
  • Ad Size
  • Ad Shown (Creative Variation)
• Less Impactful Attributes:
  • Time of Day
  • Language

Exploratory Data Analysis (EDA) using Tableau

Tableau was employed to conduct an initial exploratory analysis, allowing for manual inspection of attribute interactions. This effort spanned several months, involving:

• Visualization of attribute relationships
• Identification of trends and correlations
• Ranking of attributes by influence on click-through rates (CTR)

PCA Implementation & Automation

After manually identifying key variables, we automated the process of discovering the relative importance of each component through PCA. Key steps in the PCA process included:

1. Standardizing Data: Normalization of all numerical attributes to ensure fair weighting.
2. Computing Covariance Matrix: Evaluating how variables co-vary with one another.
3. Eigenvalue and Eigenvector Analysis: Identifying principal components that capture the most variance.
4. Dimensionality Reduction: Selecting a subset of attributes that contribute the most to click variance.
5. Weight Assignment: Automating the determination of each attribute’s weight in influencing clicks.

Findings & Insights

• Most Influential Factors:
  • Country emerged as the strongest predictor of clicks, followed by Ad Unit and Publisher/App Name.
  • Ad Size and Creative Variation also played significant roles.
• Lesser Impactful Factors:
  • Time of Day and City had a relatively minor influence on CTR.
  • Language variations contributed insignificantly to differences in engagement.

Impact & Automation Benefits

By automating PCA-based weighting, we significantly reduced the manual effort required for analyzing new data. The system now continuously evaluates ad performance, dynamically adjusting spend allocations based on empirical insights.