Court Backlog Predictor
A machine learning pipeline built on Databricks to analyze and predict judicial backlog trends across various courts.
The Problem
Court systems are overwhelmed, and administrative bodies lack predictive insights to allocate resources or judges efficiently.
The Solution
Leveraged PySpark to process historical case data and train regression models to forecast future backlog severity by case type and jurisdiction.
Architecture
Data ingested into Delta Lake, managed via Unity Catalog. PySpark handles distributed data processing and model training, with MLflow tracking experiments.
Engineering Challenges
Handling messy, unstructured historical legal data required extensive cleaning and normalization before feature engineering could begin.
Results & Impact
Successfully built and deployed a predictive pipeline during the Databricks Hackathon.
Lessons Learned
Gained hands-on experience with the modern data lakehouse architecture and distributed ML workflows.