🚲 Mi Bici Trip Duration Prediction Project

This project focuses on predicting trip durations for the "Mi Bici" public bike-sharing system in Guadalajara, Mexico. We leverage machine learning models to analyze ride patterns and provide real-time trip duration predictions.

🔗 Data Source: Mi Bici Open Data
📅 Training Data: 2024 trip records
📅 Testing Data: 2025 trip records

---

📌 Project Overview

This project integrates:

• MLflow for experiment tracking & model versioning.
• FastAPI for real-time trip duration predictions.
• Automated retraining when model performance degrades.

📂 Project Structure

.
├───api
│   ├───logs                  # Stores logs from API requests
│   ├───saved_inputs          # Stores API input data for retraining
│   │
│   │   app.py                # FastAPI application
│   │   README.md             # API documentation
│   │   ss_api_test.JPG       # Screenshot of Postman test
│
├───configs
│   │   model1.yaml           # Configuration file for model settings
│
├───data
│   ├───external              # External data sources (if applicable)
│   ├───interim               # Intermediate processed data
│   ├───processed
│   │   ├───2024              # Processed dataset for training
│   │   │   combined_2024.csv  # Combined processed data
│   │   │   test.csv          # Test dataset
│   │   │   train.csv         # Training dataset
│   │
│   └───raw
│       ├───2024              # Raw dataset from Mi Bici 2024
│       │   datos_abiertos_2024_01.csv
│       │   datos_abiertos_2024_02.csv
│       │   datos_abiertos_2024_03.csv
│       │   ...
│       │   datos_abiertos_2024_12.csv
│       │
│       │   nomenclatura_2024_12.csv  # Data dictionary
│
├───docs                      # Project documentation
│
├───MLFlow
│   ├───mlartifacts           # Stores MLflow artifacts
│   ├───mlruns                # MLflow experiment tracking
│   │
│   │   infer.py              # Model inference script
│   │   README.md             # MLflow documentation
│   │   requirements.txt      # Dependencies for MLflow
│   │   retrain_model.py      # Automated retraining script
│   │   ss_mlfow_dashboard.JPG # Screenshot of MLflow dashboard
│   │   train.py              # Model training script
│
├───models
│   │   lr_mae-286.9239_2025-01-29.pkl  # Latest linear regression model
│   │   random_forest_mae-273.4929_2025-01-29.pkl  # Latest random forest model
│
├───notebooks
│   │   1.IRA_data_preprocessing.ipynb   # Data cleaning notebook
│   │   2.IRA_data_vizualization.ipynb   # Exploratory Data Analysis (EDA)
│   │   3.IRA_modeling.ipynb             # Model training and evaluation
│   │   4.IRA_retraining.ipynb           # Model retraining analysis
│
├───references                 # References and additional documentation
│
├───reports
│   ├───figures
│   │   umap_2d_K3.png        # Visualization of UMAP embeddings
│
└───src
    ├───data
    │   │   build_features.py # Feature engineering
    │   │   cleaning.py       # Data cleaning functions
    │   │   ingestion.py      # Data loading
    │   │   labeling.py       # Labeling for supervised learning
    │   │   splitting.py      # Train-test split
    │   │   validation.py     # Data validation
    │
    ├───models
    │   ├───model1
    │   │   │   dataloader.py  # Loads data for training
    │   │   │   hyperparameters_tuning.py # Hyperparameter tuning
    │   │   │   model.py       # Model definition
    │   │   │   predict.py     # Inference script
    │   │   │   preprocessing.py # Preprocessing functions
    │   │   │   train.py       # Training script
    │
    ├───visualization
    │   │   evaluation.py     # Model evaluation scripts
    │   │   exploration.py    # Data exploration and visualization
    │
    │   __init__.py

.env  
.gitignore  
LICENSE  
Makefile  
README.md  
requirements.txt  

---

1️⃣ Model Training & Logging (MLflow)

• Uses Linear Regression to predict trip durations.
• Trained on 2024 Mi Bici data.
• Logs:
  • Hyperparameters
  • Performance metrics (MAE, RMSE)
  • Model artifacts & schema in MLflow.

2️⃣ Real-time Predictions (FastAPI)

• Deploys the trained model as a REST API.
• Logs every request & prediction for monitoring.
• Saves input data (saved_inputs/prediction_inputs.csv) for future retraining.

3️⃣ Automated Model Retraining

• Monitors model performance on 2025 Mi Bici test data.
• Retrains the model if MAE increases beyond a threshold.
• Uses both stored API inputs & new Mi Bici data for retraining.
• Registers new models in MLflow if performance improves.

---

🛠 Technologies Used

• Python (ML model + API)
• scikit-learn (Linear Regression + preprocessing)
• FastAPI (Real-time model inference)
• MLflow (Model tracking & versioning)
• pandas & NumPy (Data processing)
• joblib (Model serialization)
• Logging & CSV storage (For monitoring & retraining)

---

🚀 How It Works

1️⃣ Train the Model (train.py):

• Loads Mi Bici 2024 dataset → Preprocesses → Trains model → Logs in MLflow.

2️⃣ Deploy API (api.py):

• Loads MLflow model → Serves predictions via FastAPI.

3️⃣ Log Predictions for Retraining:

• API logs input requests → Saves them for future training.

4️⃣ Monitor & Retrain (retrain_model.py):

• Uses Mi Bici 2025 test data to check performance.
• If MAE degrades, merges new inputs + original data.
• Retrains & registers new model in MLflow.

---

📈 Data Overview

Mi Bici Dataset (2024-2025)

Column Name	Description
Trip_Id	Unique trip identifier
User_Id	Unique user identifier
Gender	Gender of the user
Year_of_Birth	Year of birth of the user
Trip_Start	Start timestamp of the trip
Trip_End	End timestamp of the trip
Origin_Id	Origin bike station ID
Destination_Id	Destination bike station ID
Trip_Duration	Total trip duration in seconds
Start_Hour	Hour of day when the trip started
Start_DayOfWeek	Day of the week (Monday=0, Sunday=6)

---

📌 Running the Project

1️⃣ Install Dependencies

pip install -r requirements.txt

2️⃣ Train & Log the Model

python scripts/train.py

This will train the model using Mi Bici 2024 data and log it in MLflow.

3️⃣ Run the FastAPI Server

uvicorn scripts.api:app --host 0.0.0.0 --port 8000 --reload

The API will be available at http://127.0.0.1:8000.

4️⃣ Make a Prediction

Using PowerShell

$headers = @{
    "Content-Type" = "application/json"
}

$body = @{
    "Year_of_Birth" = 1995
    "Gender" = 1
    "Origin_Id" = 10
    "Destination_Id" = 50
    "Start_Hour" = 14
    "Start_DayOfWeek" = 2
} | ConvertTo-Json -Depth 10

$response = Invoke-RestMethod -Uri "http://127.0.0.1:8000/predict" -Method Post -Headers $headers -Body $body
Write-Output $response

Using Curl

curl -X POST "http://127.0.0.1:8000/predict" \
     -H "Content-Type: application/json" \
     -d '{"Year_of_Birth": 1995, "Gender": 1, "Origin_Id": 10, "Destination_Id": 50, "Start_Hour": 14, "Start_DayOfWeek": 2}'

5️⃣ Monitor MLflow

mlflow ui

Check experiment logs & model versions at http://127.0.0.1:5000.

6️⃣ Schedule Automated Retraining

Schedule retrain_model.py to run daily (or as needed) using:

• Linux (Cron Job)

  crontab -e

  Add:

  0 */12 * * * /usr/bin/python3 /path/to/retrain_model.py

• Windows (Task Scheduler)
  • Set Trigger: Run every 12 hours.
  • Set Action: Run:

    python C:\path\to\retrain_model.py