📊 Live Dashboard: Launch the App
🧠 Project Type: End-to-End Data Science | ML Modeling | Streamlit Dashboard | Real-World Analysis
This project aims to predict on-time delivery for Domino’s orders using machine learning and uncover the key drivers that influence delivery success. The solution includes an interactive Streamlit dashboard, model-driven insights, and post-launch performance analysis — even when the results were unexpected.
On-time delivery is critical for customer satisfaction, revenue, and brand loyalty in food delivery. This project tackles:
- Predicting whether an order will be delivered on time
- Understanding which factors (weather, order size, traffic) drive that outcome
- Comparing pre-test vs post-test performance to evaluate real business impact
| Layer | Tools & Libraries |
|---|---|
| Data | CSV (Dominos_Full_Example.csv) |
| Modeling | Python, Pandas, Scikit-learn, RandomForest, LogisticRegression |
| Dashboard | Streamlit, Plotly, Statsmodels |
| Visualization | Plotly Express, Graph Objects |
| Deployment | Streamlit Cloud |
The live dashboard includes:
- ✅ Pre vs Post Test on-time delivery comparison
- ✅ Actual vs Predicted delivery performance by test phase
- ✅ Model accuracy + business metrics
- ✅ Correlation matrix and variable impact (Order Size, Weather, Distance)
- ✅ Store-location and order-type performance heatmaps
- ✅ Dynamic filters (Test Phase, Store, Order Type)
Two models were trained:
- Logistic Regression for interpretability
- Random Forest for performance
| Metric | Logistic Regression | Random Forest |
|---|---|---|
| Accuracy | 82.09% | 84.77% |
| Precision | 85.99% | 88.59% |
| Recall | 86.45% | 87.82% |
| F1 Score | 86.22% | 88.20% |
📌 Key Drivers Identified by the Model:
- High impact: Order Size, Total Dollars
- Negative impact: Weather, Traffic
- Unexpected insight: Distance & Peak Hours had minimal effect
Despite strong model performance, the real-world post-test delivery performance dropped by 31.3%.
- On-time delivery rate fell from ~50% to ~18%
- This was visualized clearly in the dashboard, with metric cards and comparison charts
- The model predicted well on validation data but did not translate into business improvement
- Possible causes:
- External disruptions (e.g., weather, staffing, volume spike)
- Poor operational execution of the model insights
- Data drift or mismatch between pre-test and post-test conditions
A model can be statistically strong, but if not implemented well, the real-world impact can still fail.
This project shows not just ML skill — but the ability to diagnose performance gaps through dashboards and business thinking.
- Prioritize high-value, high-complexity orders (they’re more likely to be on time)
- Re-evaluate the operational rollout plan post-model deployment
- Investigate weather and traffic effects further — consider real-time route planning
- Monitor model vs actual results continuously via dashboards
dominos-delivery-performance/
├── dominos_dashboard.py # Streamlit dashboard code
├── dominos_model.ipynb # Modeling + EDA notebook
├── Dominos_Full_Example.csv # Dataset
├── dominos_report.pdf # Business-style project report
├── requirements.txt # For deployment
├── README.md # You're reading this!
Aadira Anil Ramakrishnan
MS in Analytics, Northeastern University
LinkedIn
MIT License. See LICENSE file for details.