BikeML-API: Analyzing the Behavior of Bike-Sharing Users in Montreal

---

Contents

• 1. Overview
• 2. Project Goals
• 3. Project Features
• 4. Tools and Frameworks
• 5. Directory Structure
• 6. Reproducibility
• 7. How to Use the API
• 8. Data Sources

---

1. Overview

The BikeML-API explores the relationship between weather conditions and bike-sharing usage in Montreal. It includes a machine learning pipeline to predict trip durations based on weather and temporal features, as well as an API for serving predictions. The findings aim to help bike-sharing operators like BIXI optimize operations and user satisfaction.

---

2. Project Goals

The goals of this project are:

1. To analyze how weather conditions affect the duration of bike-sharing trips.
2. To build and deploy a machine learning model that predicts trip durations.
3. To provide actionable insights for bike-sharing companies to optimize bike availability.

---

3. Project Features

• Weather and Trip Data Integration: Combining weather observations with BIXI trip data for comprehensive analysis.
• Machine Learning Model: A Random Forest model trained to predict trip durations.
• API Deployment: A Flask API for serving predictions based on user-input weather and temporal features.

---

4. Tools and Frameworks

Tools:

• Python: Data preprocessing, model training, and API development.
• Jupyter Notebooks: For exploratory data analysis (EDA) and feature engineering.

Libraries:

• Scikit-learn: Model development and evaluation.
• Flask: Deployment of the prediction API.
• Pandas/NumPy: Data manipulation.
• Matplotlib/Seaborn: Visualization.

Other Tools:

• Git LFS: For tracking large files in the repository.
• Docker (Optional): For containerizing the API.

---

5. Directory Structure

BikeML-API/
│
├── data/
│   ├── bixi_data_2021.csv                      # BIXI bike-sharing trip data (Git LFS tracked)
│   ├── weather_data_2021.csv                   # Weather data for Montreal (Git LFS tracked)
│   ├── refined_combined_data_with_features.xls # Final processed dataset (Git LFS tracked)
│   └── ...
│
├── imgs/                                       # (Optional) Visualization images
│   └── ...
│
├── models/
│   ├── bike_duration_predictor.pkl             # Trained model for predictions
│   ├── tuned_random_forest.pkl                 # Tuned Random Forest model
│   └── ...
│
├── notebooks/
│   ├── BikeML01.ipynb                          # Data preprocessing and feature engineering
│   ├── BikeML02.ipynb                          # Model training and evaluation
│   └── ...
│
├── app.py                                      # Flask API script
├── predict_request.ps1                         # Script for making POST requests to the API
├── requirements.txt                            # Python dependencies
├── Dockerfile                                  # Dockerfile for containerization (optional)
└── README.md                                   # Project description and instructions

plaintext

---

6. Reproducibility

Steps to Reproduce:

1. Clone this repository:

   git clone https://github.com/AFARNOOD/BikeML-API.git
   cd BikeML-API
   

2. Set up Python dependencies:

   pip install -r requirements.txt
   

3. Download the required datasets or ensure they are placed in the data/ directory.
4. Run the Jupyter Notebooks in the notebooks/ directory to preprocess the data or retrain the model.

7. How to Use the API

To use the API for predictions, follow these steps:

1. Run the Flask App: Ensure the Flask API is running locally. Start the API by executing the following command in your terminal:

   python app.py

   The server will run on http://127.0.0.1:5000.

2. Send a POST Request: Use the provided predict_request.ps1 script or your preferred tool (e.g., Postman, cURL) to send a request with input data for prediction. The input JSON must include the following features:

   {
       "max_temp_c": 20,
       "min_temp_c": 7,
       "temp_range_c": 18,
       "total_precip_mm": 2,
       "snow_on_grnd_cm": 0,
       "spd_of_max_gust_kmh": 10,
       "is_rainy": 1,
       "is_snowy": 0,
       "is_windy": 0,
       "start_hour": 9,
       "start_weekday": 1,
       "is_weekend": 0
   }

3. Example PowerShell Command: Run the following command in PowerShell to send the prediction request:

   Invoke-RestMethod -Uri http://127.0.0.1:5000/predict `
       -Method POST `
       -ContentType "application/json" `
       -Body '{"max_temp_c": 20, "min_temp_c": 7, "temp_range_c": 18, "total_precip_mm": 2, "snow_on_grnd_cm": 0, "spd_of_max_gust_kmh": 10, "is_rainy": 1, "is_snowy": 0, "is_windy": 0, "start_hour": 9, "start_weekday": 1, "is_weekend": 0}'

4. Example cURL Command: If you prefer using cURL, run this command in your terminal:

   curl -X POST http://127.0.0.1:5000/predict \
     -H "Content-Type: application/json" \
     -d '{"max_temp_c": 20, "min_temp_c": 7, "temp_range_c": 18, "total_precip_mm": 2, "snow_on_grnd_cm": 0, "spd_of_max_gust_kmh": 10, "is_rainy": 1, "is_snowy": 0, "is_windy": 0, "start_hour": 9, "start_weekday": 1, "is_weekend": 0}'
   

5. Expected Output: The API will respond with the predicted trip duration in seconds. Example response:

   {
    "predicted_duration_sec": 789
   }

Note: For other tools or environments, ensure the JSON request body matches the required format and is sent as a POST request to the /predict endpoint.

---

8. Data Sources

This project uses two primary datasets:

1. Weather Data:

   • Source: Environment and Climate Change Canada (ECCC).
   • Description: Includes daily and hourly weather data for Montreal in 2021.
   • Features: Temperature, precipitation, wind speed, and other climatic conditions.

   Download Weather Data

2. BIXI Data:

   • Source: BIXI Montreal Open Data Portal.
   • Description: Contains detailed trip records of BIXI bike-sharing users for 2021.
   • Features: Trip start and end times, station locations, and duration.

   Download BIXI Data