Predicting Customer Satisfaction for Future Purchases (MLOps)

🚀 Project Overview

This project aims to predict the customer satisfaction score for the next order or purchase based on a customer's historical data. I use the Brazilian E-Commerce Public Dataset by Olist to analyze 100,000+ orders across various marketplaces from 2016 to 2018. By predicting satisfaction scores from features like order status, price, and reviews, the project demonstrates how to build production-ready ML pipelines using ZenML and MLflow.

The pipeline trains a machine learning model that predicts customer satisfaction and deploys it using MLflow, enabling continuous monitoring and deployment.

📄 Problem Statement

Given a customer's order history, we predict the review score for their next purchase based on features such as:

• Order status
• Price
• Payment method
• Freight performance
• Customer reviews

I create a ZenML pipeline that trains, evaluates, and deploys a machine learning model to predict customer satisfaction. By integrating MLflow, we can track model performance and ensure continuous deployment for future predictions.

🔧 Tools & Frameworks

• ZenML: Framework for building and deploying production-grade ML pipelines.
• MLflow: Tool for tracking experiments and model deployment.
• Python 3.8+: Core programming language for model development.

🧠 Project Workflow

1. Data Ingestion: Load and preprocess data from the Brazilian e-commerce dataset.
2. Data Cleaning: Remove unwanted columns and outliers.
3. Model Training: Train a machine learning model and log results using MLflow autologging.
4. Model Evaluation: Evaluate the model's performance based on the Mean Squared Error (MSE).
5. Model Deployment: Deploy the model using MLflow, enabling real-time predictions.
6. Continuous Deployment Pipeline: Automate retraining and redeployment when model accuracy improves beyond a set threshold.

🛠 How to Run

Follow the steps below to clone and set up the repository:

# Clone the repository
git clone https://github.com/Mayankpratapsingh022/Customer-Satisfaction-MLOps.git

# Navigate to the project folder
cd Customer-Satisfaction-MLOps

# Install dependencies
pip install -r requirements.txt

# Set up ZenML with MLflow
pip install zenml[server]
zenml up
zenml integration install mlflow -y

📈 Training Pipeline

The training pipeline consists of several steps:

• Ingest Data: Load the dataset and create a pandas DataFrame.
• Clean Data: Perform data cleaning and feature selection.
• Train Model: Train a machine learning model and log performance with MLflow autologging.
• Evaluate Model: Calculate metrics such as MSE to evaluate the model.

🚀 Deployment Pipeline

The deployment pipeline handles continuous model deployment with MLflow:

1. Deployment Trigger: Check if the newly trained model meets accuracy requirements.
2. Model Deployer: Deploy the model if the evaluation metric passes the threshold.

The pipeline automatically updates the MLflow deployment server with the latest model version when performance improves.

📚 Resources & References

• Dataset: Brazilian E-Commerce Public Dataset by Olist
• ZenML Documentation: ZenML
• MLflow Documentation: MLflow

🏆 Results

Using this framework, you can continuously predict customer satisfaction, deploy models in production, and monitor the performance for future improvements.