Digital Content Consumption Patterns

Customer Analytics Portfolio — Unsupervised Learning

End-to-end pipeline that generates synthetic session-level data from a digital content platform and discovers behavioral consumption segments ("digital personas") using dimensionality reduction (PCA / t-SNE) and clustering (K-Means, with DBSCAN diagnostics).

---

Project Overview

The project simulates user sessions and identifies four latent behavioral clusters:

Persona	Behavioral Signature
Marathon Finishers	Long sessions, near-complete depth, high interactions, desktop, daytime
Sampling Nomads	Many pieces, very low depth, high diversity, mobile, afternoon/evening
Burst Consumers	Short intense sessions, moderate depth, mobile-only, sporadic timing
Download & Depart	Very brief sessions, high save/download actions, minimal in-session depth

---

Project Structure

The project follows a modular and reproducible structure:

.
├── data/
│   ├── raw/                  # Synthetic dataset, data dictionary, cluster definitions
│   └── processed/            # Scaled matrix, PCA/t-SNE outputs, cluster assignments
│
├── notebooks/
│   ├── 01_exploratory_analysis.ipynb
│   ├── 02_clustering_modeling.ipynb
│   └── 03_evaluation_insights.ipynb
│
├── src/
│   ├── __init__.py
│   ├── config.py             # Centralized paths, columns, and constants
│   ├── data_generation.py    # Phase 0: Synthetic data generation
│   ├── eda.py                # Phase 1: Exploratory analysis
│   ├── preprocessing.py      # Phase 2: Scaling and encoding
│   ├── dimensionality_reduction.py  # Phase 3: PCA and t-SNE
│   ├── clustering.py         # Phase 4: K-Means clustering
│   ├── profiling.py          # Phase 5: Cluster profiling and persona mapping
│   └── pipeline.py           # Orchestrates all phases
│
├── models/                  # Persisted models (e.g., .pkl files)
│
├── reports/
│   ├── figures/              # Visualizations: EDA, clustering diagnostics, and profiling
│   │   ├── univariate_distributions.png
│   │   ├── boxplot_outliers.png
│   │   ├── correlation_heatmap.png
│   │   ├── categorical_distributions.png
│   │   ├── pairplot_relationships.png
│   │   ├── preprocessing_scaling_impact.png
│   │   ├── pca_explained_variance.png
│   │   ├── pca_loadings_heatmap.png
│   │   ├── dual_projection_comparison.png
│   │   ├── kmeans_diagnostics.png
│   │   ├── dbscan_kdistance.png
│   │   ├── silhouette_validation.png
│   │   └── cluster_radar_profiles.png
│   │
│   └── tables/               # Summary tables: EDA, preprocessing, clustering, and profiling
│       ├── descriptive_statistics.csv
│       ├── outlier_analysis_summary.csv
│       ├── spearman_correlation_matrix.csv
│       ├── categorical_distributions.csv
│       ├── data_quality_report.csv
│       ├── scaling_report.csv
│       ├── onehot_encoding_map.csv
│       ├── final_feature_matrix_columns.csv
│       ├── pca_variance_breakdown.csv
│       ├── pca_dominant_features.csv
│       ├── kmeans_k_selection.csv
│       ├── kmeans_cluster_distribution.csv
│       ├── dbscan_grid_search.csv
│       ├── dbscan_failure_diagnosis.csv
│       ├── model_selection_comparison.csv
│       ├── silhouette_per_cluster.csv
│       ├── cluster_numerical_profiles.csv
│       ├── cluster_categorical_profiles.csv
│       ├── cluster_persona_mapping.csv
│       └── cluster_persona_detailed.csv
│
├── requirements.txt
└── README.md

Installation

1. Clone the repository:

   git clone <repository-url>
   cd <repository-folder>

2. (Optional) Create a virtual environment:

   python -m venv .venv
   source .venv/bin/activate   # On Windows: .venv\Scripts\activate

3. Install dependencies:

   pip install -r requirements.txt

---

Execution

Option 1: Full Pipeline

Run the entire pipeline in one command:

python -m src.pipeline

This executes all phases sequentially: Data Generation → EDA → Preprocessing → Dimensionality Reduction → Clustering → Profiling.

Option 2: Step-by-Step Execution

1. Generate synthetic data:

   python generate_project.py

   Outputs:

   • src/ with all phase modules
   • notebooks/ with Jupyter notebooks
   • data/raw/ with the dataset and definitions

2. Run individual phases: Open and execute the notebooks in order:

   • 01_exploratory_analysis.ipynb
   • 02_clustering_modeling.ipynb
   • 03_evaluation_insights.ipynb

---

Pipeline Phases

Phase	Module	Output
0	data_generation.py	data/raw/digital_content_sessions.csv
1	eda.py	reports/outlier_analysis_summary.csv, heatmap
2	preprocessing.py	data/processed/sessions_preprocessed.csv
3	dimensionality_reduction.py	sessions_pca.csv, sessions_tsne.csv, loadings
4	clustering.py	cluster_assignments.csv, models/kmeans_model.pkl
5	profiling.py	reports/cluster_persona_mapping.csv

---

Methodology

Data Processing

• Scaling: StandardScaler for numerical features (zero-mean, unit-variance).
• Encoding: One-hot encoding for categorical features.

Dimensionality Reduction

• PCA: Retains ≥80% of variance for clustering input.
• t-SNE: Used for visualization only (not for clustering).

Clustering

• K-Means: Optimal k selected via silhouette score, validated with Davies–Bouldin index.
• DBSCAN: Included for diagnostic comparison.

Reproducibility

• Global RANDOM_STATE = 42 for consistent results.

---

Notes

• All scripts are designed to read/write files locally. Execute in a local environment (terminal, VS Code, Jupyter).
• t-SNE requires scikit-learn ≥ 1.2 due to the max_iter parameter.

---

License

This project is intended for portfolio and educational purposes.