https://github.com/bhawnamehbubani/food-delivery-time-prediction-model

A cutting-edge food delivery time prediction model leveraging Random Forest Regressor, Streamlit, OpenCage API, GeoPy, and advanced feature engineering techniques to optimize delivery operations and customer satisfaction.
https://github.com/bhawnamehbubani/food-delivery-time-prediction-model

feature-engineering geopy opencage-api random-forest-regressor streamlit

Last synced: 4 months ago
JSON representation

A cutting-edge food delivery time prediction model leveraging Random Forest Regressor, Streamlit, OpenCage API, GeoPy, and advanced feature engineering techniques to optimize delivery operations and customer satisfaction.

• Host: GitHub
• URL: https://github.com/bhawnamehbubani/food-delivery-time-prediction-model
• Owner: BhawnaMehbubani
• License: mit
• Created: 2025-01-25T08:42:52.000Z (9 months ago)
• Default Branch: main
• Last Pushed: 2025-01-25T09:51:10.000Z (9 months ago)
• Last Synced: 2025-02-06T03:52:28.030Z (9 months ago)
• Topics: feature-engineering, geopy, opencage-api, random-forest-regressor, streamlit
• Language: Jupyter Notebook
• Homepage:
• Size: 2.38 MB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
# **Food Delivery Time Prediction Model**  

## **Overview**  

This project delivers a cutting-edge **food delivery time prediction model** designed to enhance delivery logistics and optimize customer satisfaction. By leveraging state-of-the-art machine learning algorithms, advanced geospatial analysis, and a user-friendly deployment interface, it offers both precise predictions and actionable insights.  

**Tech Stack Highlights**:  

- **Machine Learning**: Random Forest Regressor, XGBoost for predictive modeling.  

- **Feature Engineering**: Advanced techniques for geospatial and temporal features.  

- **API Integration**: OpenCage API and GeoPy for precise location mapping.  

- **Frontend**: Streamlit for seamless real-time prediction and interactive visualization.  

## **Table of Contents**  

1. [Data Source](#data-source)  

2. [Pipeline Architecture](#pipeline-architecture)  

3. [Feature Engineering](#feature-engineering)  

4. [Machine Learning Model](#machine-learning-model)  

5. [Technologies and Tools](#technologies-and-tools)  

6. [Data Insights and Real-World Observations](#data-insights-and-real-world-observations)  

7. [Results and Evaluation](#results-and-evaluation)  

8. [Deployment](#deployment)  

9. [Future Improvements](#future-improvements)  

## **Data Source**  

The dataset encapsulates various real-world attributes influencing delivery times:  

- **Order Details**: Customer locations, pickup/drop-off points.  

- **Delivery Partner Metrics**: Ratings, historical performance, current workload.  

- **External Factors**: Weather conditions, traffic levels, peak hours.  

- **Delivery Outcomes**: Actual delivery times (target variable).  

## **Pipeline Architecture**  

```plaintext

Data Collection  

     |  

     v  

Data Preprocessing  

  ┌──────────────────────────────────────────────────────────────────────────────┐  

  | - Missing Value Handling (Weather, Ratings via Median Imputation)           |  

  | - Outlier Removal (Extreme Delivery Times > 180 mins due to Accidents)      |  

  | - Standardizing Numerical Features (Distance, Time)                        |  

  | - Encoding Categorical Variables (Traffic Levels, Weather Conditions)      |  

  └──────────────────────────────────────────────────────────────────────────────┘  

     |  

     v  

Feature Engineering  

  ┌──────────────────────────────────────────────────────────────────────────────┐  

  | - Geospatial Distance Calculation Using GeoPy and Haversine Formula         |  

  | - Time-Based Features (Weekends, Peak Hours)                                |  

  | - API Integration for Address Normalization Using OpenCage API              |  

  | - Correlation Analysis to Select Optimal Features                           |  

  └──────────────────────────────────────────────────────────────────────────────┘  

     |  

     v  

Model Selection and Training  

  ┌──────────────────────────────────────────────────────────────────────────────┐  

  | - Compared Algorithms:                                                     |  

  |   • XGBoost (Chosen for Accuracy, Feature Insights)                        |  

  |   • Random Forest (Robust Baseline Model)                                  |  

  |   • Linear Regression (Excluded for Poor Fit on Non-linear Data)           |  

  | - Hyperparameter Tuning via Grid Search                                    |  

  └──────────────────────────────────────────────────────────────────────────────┘  

     |  

     v  

Model Evaluation  

  ┌──────────────────────────────────────────────────────────────────────────────┐  

  | - Metrics Used:                                                            |  

  |   • R² Score: Goodness of Fit                                              |  

  |   • RMSE: Overall Prediction Accuracy                                      |  

  |   • MAE: Average Deviation in Predictions                                  |  

  └──────────────────────────────────────────────────────────────────────────────┘  

     |  

     v  

Deployment  

  ┌──────────────────────────────────────────────────────────────────────────────┐  

  | - Built with Streamlit for Real-Time Predictions and Interactive Interface  |  

  | - Model Integration (`model.pkl`)                                          |  

  └──────────────────────────────────────────────────────────────────────────────┘  

```

## **Feature Engineering**  

### **Key Features**  

1. **Distance Calculation**:  

   - **Tools Used**: GeoPy, Haversine Formula, OpenCage API for geocoding.  

   - **Insight**: Deliveries over 15 km showed higher variability due to inconsistent traffic patterns.  

2. **Time-Based Features**:  

   - **What**: Weekends, peak hours, and holidays significantly impact delivery times.  

   - **Insight**: Weekend orders were 20% faster due to lower traffic levels.  

3. **Traffic and Weather Conditions**:  

   - **Tools Used**: Encoded categories for "Clear," "Rain," and "Heavy Traffic."  

   - **Insight**: Rainy weather consistently added 9 minutes to delivery times, and heavy traffic extended it by ~12 minutes.  

4. **Delivery Partner Ratings**:  

   - **What**: Correlated partner ratings with delivery efficiency.  

   - **Insight**: Partners rated >4.5 completed deliveries 7% faster on average.  

## **Machine Learning Model**  

### **Algorithm Selection**  

1. **XGBoost (Final Model)**:  

   - Outperformed others due to its ability to handle complex, non-linear relationships.  

   - Provided interpretable feature importance scores.  

2. **Random Forest (Baseline)**:  

   - Delivered strong baseline performance but lacked the precision of XGBoost in edge cases.  

3. **Linear Regression**:  

   - Rejected due to poor performance (R² ~0.45) on non-linear features like weather and traffic.  

## **Technologies and Tools**  

| **Technology** | **Purpose** |  

|-----------------|-------------|  

| **Python**     | Core programming language for modeling and analysis. |  

| **Jupyter**    | Environment for exploratory data analysis and prototyping. |  

| **Streamlit**  | Web app for deployment and real-time prediction. |  

| **GeoPy**      | Calculated geospatial distances. |  

| **OpenCage API** | Geocoded addresses to standardize location data. |  

| **XGBoost**    | Machine learning model with the highest accuracy. |  

| **Random Forest** | Baseline model for robustness comparisons. |  

## **Data Insights and Real-World Observations**  

1. **Peak Hours Delay**:  

   - Lunch-time deliveries faced 18% longer delays due to high traffic.  

   - **Realization**: Restaurants can pre-prepare meals for peak orders.  

2. **Weather Disruptions**:  

   - Rainy days increased delivery times by ~20%, with heavy rain causing delays up to 30%.  

   - **Actionable Insight**: Incorporate weather forecasts into ETA calculations.  

3. **Short-Distance Traffic Bottlenecks**:  

   - Urban congestion caused delays for deliveries under 2 km, especially in city centers.  

   - **Recommendation**: Use bikes or two-wheelers for dense urban areas.  

4. **Delivery Partner Efficiency**:  

   - High-rated partners consistently outperformed their peers in timely deliveries.  

   - **Lesson**: Incentivize partners with higher ratings during peak periods.  

## **Results and Evaluation**  

- **Best Model**: XGBoost  

- **Performance Metrics**:  

  - **R² Score**: 0.82  

  - **RMSE**: 8.23 minutes  

  - **MAE**: 6.45 minutes  

## **Deployment**  

**Streamlit Application**:  

- Deployed a real-time interface for predictions.  

- Allows users to input features (distance, weather, traffic) and instantly receive delivery time estimates.  

## **Future Improvements**  

1. **Real-Time Traffic API**: Integrate live traffic data for dynamic predictions.  

2. **Weather Severity Analysis**: Include detailed weather data like storm warnings.  

3. **Feature Expansion**: Add features like restaurant preparation times.  

4. **Advanced Models**: Experiment with ensemble models combining XGBoost and neural networks.  

## **Contributions**  

Contributions are welcome! Feel free to fork the repository, make improvements, and submit a pull request.