Supreme Court decision prediction has been an area of interest since the creation of the highest court. Despite interest in the area, both for intellectual and business reasons, nobody has found a consistent way to do so accurately. Experts for the most part barely outperform the baseline of ~2/3 accuracy.

This work improves upon prior art to increase automated Supreme Court decision classification accuracy. In particular, it focuses on two primary areas of improvement on previous methods:

1. Featurization complexity: Prior art relies on a single source of data: either unstructured court hearing transcripts or structured information about historical court cases and outcomes in database format (most commonly SCDB). This work combines both types of information by extracting complex textual features from hearing transcripts and joining them with structured features derived from SCDB.
2. Machine learning model: Prior art leverages simple machine learning models, most commonly random forests. This works introduces more complex machine learning models and experiments with variety of model, hyperparameter combinations to empirically identify those best suited for this task.

The resulting system demonstrates accuracy that exceeds the state of the art by approximately 2%.

Project Leader: Ashkon Farhangi, former CodeX fellow

The descriptions of current and past projects of CodeX non-residential fellows are provided to illustrate the kind of work our non-residential fellows are carrying out. These projects are listed here for informational purposes only and are not endorsed by CodeX, Stanford Law School, or Stanford University.