IoT Traffic Volume Prediction with MLOps

This project demonstrates a full MLOps workflow for predicting traffic volume using an XGBoost regression model. Built with ZenML, the pipeline includes steps for data ingestion, preprocessing, training, evaluation, and deployment using MLflow.

---

Problem Statement

Predict the number of vehicles using historical IoT traffic sensor data with features like Hour, Weekday, Day, Month, and Week.

---

🛠️ Tech Stack

• ZenML – For pipeline orchestration
• XGBoost – Regressor model
• MLflow – Model deployment and tracking
• Docker – Containerized pipeline steps
• pandas / scikit-learn – Data handling and preprocessing

---

⚙️ Pipelines

1. Training Pipeline

Trains and evaluates the model.

Steps:

• Load data
• Clean/split data
• Train model (XGBRegressor)
• Evaluate model (RMSE)

training_pipeline(data_path="data/train_ML_IOT.csv")

---

2. Continuous Deployment Pipeline

Trains a model and deploys it via MLflow if RMSE < max_rmse.

Steps:

• Ingest and clean data
• Train and evaluate model
• Trigger deployment based on RMSE
• Deploy model using MLflow

continuous_deployment_pipeline(max_rmse=50.0)

---

🐳 Docker Integration

Each step in the pipeline runs in a containerized environment using ZenML's DockerSettings.

---

Getting Started

1. Clone and Install

git clone https://github.com/lokk798/iot-traffic-mlops.git
cd iot-traffic-mlops
pip install -r requirements.txt

2. Initialize ZenML Repository

zenml init
zenml login --local

3. Register Stack Components

zenml integration install mlflow -y
zenml experiment-tracker register mlflow_tracker_customer --flavor=mlflow
zenml model-deployer register mlflow_customer --flavor=mlflow
zenml stack register mlflow_stack_customer -a default -o default -d mlflow_customer -e mlflow_tracker_customer --set

4. Run Pipelines

python run_training.py
python run_deployment.py

---

📌 Future Improvements

• Monitor drift in deployed models.
• Add CI/CD via GitHub Actions.

---

Author

Loukmane Daoudi – LinkedIn

---

📄 License

This project is licensed under the MIT License.