Predictive Analytics with Python (en Inglés) - Rahul Kumar Thatikonda

Build Models That Survive Beyond the Notebook.

Book Description

Data Science Finds the Signal. Engineering Turns It into Business Value.

Moving a model from a Jupyter Notebook to a production system requires engineering discipline, not just data science skills. Predictive Analytics with Python is the definitive guide for the engineering-first era of data science, helping you transition from fragile notebook workflows to resilient, production-ready predictive systems built for real-world infrastructure.

You begin by replacing slow legacy workflows with a modern technical stack, high-performance ETL with Polars, data contract enforcement with Pandera, and resilient Scikit-Learn and XGBoost pipelines with rigorous feature engineering, cross-validation, and experiment tracking using MLflow. The book then advances into time-series forecasting with Nixtla before covering model serialisation, REST API deployment with FastAPI, Docker containerisation, and production monitoring as well as governance.

The book culminates in an end-to-end capstone project building an enterprise-grade Automated Real Estate Valuation Model. By the end, you will engineer predictive systems that prioritize stability, auditability, and transformative business value.

What you will learn

● Transition fragile notebook workflows into robust production-grade software engineering practices.

● Execute high-performance ETL and data processing using the Polars library at scale.

● Enforce rigorous data contracts using Pandera to validate pipeline inputs automatically.

Table of Contents

1. From Notebooks to Systems

2. The Modern Python Environment

3. High-Performance ETL with Polars

4. Defensive Data Programming with Pandera

5. Feature Engineering as Software

6. Handling Real-World Messiness

7. The Baseline: Linear Pipelines

8. Productionizing Gradient Boosting (XGBoost)

9. The Tuning Lifecycle and Experiment Tracking

10. Model Evaluation and Interpretation

11. Engineering Time-Series Features

12. Modern Forecasting with Nixtla

13. The Deployment Gap: Serialization and Packaging

14. Serving Predictions with APIs

15. Monitoring and Model Governance

16. Capstone: Building the Enterprise AVM

       Index