MetricFlow — Social Media Campaign Analytics

End-to-end ML pipeline analyzing 200,000 ad campaigns across 6 channels — from raw CSV to predictive intelligence.

---

Overview

MarketPulse ingests 200,000 social media ad campaigns spanning 6 channels (Email, Facebook, Google Ads, Instagram, Website, YouTube), runs them through a 10-stage ML pipeline, and surfaces insights via 14 static charts, 4 interactive Plotly dashboards, and a live React web app.

Dataset	200,000 campaigns · 16 raw features → 24+ engineered
Revenue analyzed	$12.52B total across all channels
ML models	5 classifiers benchmarked (best: XGBoost 87.3% accuracy)
Dashboards	4 interactive Plotly HTML dashboards
Charts	14 Matplotlib/Seaborn static visualizations
UI	React 18 + Tailwind CSS · deployed on Vercel

---

Live Demo

https://metricflow-analytics.vercel.app

The React UI showcases every output of the pipeline — KPI cards, animated bar charts, a filterable chart gallery with modal zoom, ML model scorecards, and links to the 4 live Plotly dashboards.

---

Pipeline Architecture

Social_Media_Advertising.csv  (200,000 rows)
          │
          ▼
  ┌───────────────┐
  │  P1 — Clean   │  Fix Duration format · parse Acquisition_Cost
  │               │  Derive CTR, CPC, CPM, Revenue, Profit, ROI
  └──────┬────────┘
         │
  ┌──────▼────────┐
  │  P2–P4 — EDA  │  Channel KPIs · Audience Segments · A/B tests
  │  & Stats      │  T-tests · ANOVA · Pearson correlation matrix
  └──────┬────────┘
         │
  ┌──────▼────────┐
  │  P5 — Feature │  Label encoding · StandardScaler
  │  Engineering  │  Interaction features: CTR×Conversion, ROI×Engagement
  └──────┬────────┘
         │
  ┌──────▼────────┐
  │  P6/P6b — ML  │  Logistic Regression · Random Forest
  │  Baseline     │  Gradient Boosting · 5-fold CV · leakage-free split
  └──────┬────────┘
         │
  ┌──────▼────────┐
  │  P7 — Boost   │  XGBoost · AdaBoost · Feature Importance
  └──────┬────────┘
         │
  ┌──────▼────────┐
  │  P8 — Charts  │  14 Matplotlib/Seaborn PNGs
  └──────┬────────┘
         │
  ┌──────▼────────┐
  │ P9 — Dashboards│ 4 Plotly interactive HTML dashboards
  └──────┬────────┘
         │
  ┌──────▼────────┐
  │ P10 — Master  │  Single call executes the full pipeline end-to-end
  └───────────────┘

---

ML Model Results

Model	Accuracy	Precision	Recall	F1	AUC-ROC
XGBoost ⭐	87.3%	86.9%	87.8%	87.3%	0.94
Gradient Boosting	86.1%	85.7%	86.5%	86.1%	0.93
Random Forest	84.6%	84.2%	85.1%	84.6%	0.92
AdaBoost	81.4%	80.9%	81.8%	81.3%	0.89
Logistic Regression	76.2%	75.8%	76.7%	76.2%	0.84

Target: high-ROI campaign classification (binary). 80K sample · 5-fold stratified CV · no data leakage.

---

Channel Performance

Channel	Revenue	ROI	CTR
Facebook	$2.10B	5.02x	14.05%
Email	$2.10B	5.00x	14.05%
Google Ads	$2.10B	5.00x	13.92%
Website	$2.09B	5.01x	14.10%
YouTube	$2.08B	4.99x	14.12%
Instagram	$2.08B	4.99x	14.00%

---

Interactive Dashboards

Dashboard	Description
Revenue & ROI	Revenue by channel, goal, duration, location
Platform Insights	CTR vs spend heatmaps across all channels
Audience Segments	Segment KPIs, demographic breakdown, location ROI
ML Models	ROC curves, confusion matrix, feature importance

---

Quick Start

Requirements: Python 3.11+ · Git · Kaggle account

# 1. Clone
git clone https://github.com/devxrachit/Campaign-Optimization-Platform.git
cd Campaign-Optimization-Platform

# 2. Install dependencies
pip install -r requirements.txt

# 3. Download dataset from Kaggle
#    Search: "Marketing Campaign Performance Dataset" by manishabhatt22
#    Rename the file → Social_Media_Advertising.csv
#    Place it in:  data/Social_Media_Advertising.csv

# 4. Run the notebook
jupyter notebook notebook/Social-ads-pipeline-mergefile.ipynb
# → Run all cells, or scroll to the P10 cell and run it once to execute
#   the full pipeline end-to-end automatically

# 5. View dashboards (no notebook needed)
#    Open any HTML file directly in your browser:
#    p9_01_money_dashboard.html
#    p9_03_customers_dashboard.html
#    p9_04_models_dashboard.html

To run the UI locally:

cd ui
npm install
npm run dev        # → http://localhost:5173

---

Folder Structure

Campaign-Optimization-Platform/
│
├── notebook/
│   └── Social-ads-pipeline-mergefile.ipynb   # Full 10-stage pipeline
│
├── Matplot-charts/                            # 14 static PNG charts (p8)
│   ├── p8_01_kpi_summary.png
│   ├── p8_02_channel.png
│   ├── p8_03_pinterest.png
│   ├── p8_04_monthly.png
│   ├── p8_05_audience.png
│   ├── p8_06_campaign_goal.png
│   ├── p8_07_segment.png
│   ├── p8_08_correlation.png
│   ├── p8_09_roc_curves.png
│   ├── p8_10_model_comparison.png
│   ├── p8_11_feature_importance.png
│   ├── p8_12_confusion.png
│   ├── p8_13_duration.png
│   └── p8_14_location.png
│
├── p9_01_money_dashboard.html                 # Interactive Plotly dashboards
├── p9_02_pinterest_dashboard.html
├── p9_03_customers_dashboard.html
├── p9_04_models_dashboard.html
│
├── ui/                                        # React web app (Vite + Tailwind)
│   ├── src/
│   │   ├── components/                        # 8 page sections
│   │   └── App.jsx
│   ├── public/charts/                         # Chart PNGs served statically
│   └── vercel.json
│
├── data/                                      # Place Social_Media_Advertising.csv here
├── All Csv Files/                             # Intermediate CSVs (gitignored)
├── requirements.txt
└── README.md

---

Tech Stack

Layer	Tools
Language	Python 3.11
Data	Pandas 2.0, NumPy
Stats	SciPy (t-test, ANOVA, chi-square)
ML	Scikit-learn, XGBoost, AdaBoost
Visualization	Matplotlib, Seaborn, Plotly
UI	React 18, Vite 5, Tailwind CSS, Lucide
Deployment	Vercel
Runtime	Jupyter Notebook

---

Deployment

The React UI is deployed on Vercel and rebuilds automatically on every push to main.

Deploy your own fork:

1. Fork this repo on GitHub
2. Go to vercel.com → New Project → Import your fork
3. Set Root Directory → ui
4. Leave Build Command and Output Directory as auto-detected
5. Click Deploy

---

Made by devxrachit · Live Demo · View on GitHub