The inspiration for this project came from lesson 8 of Prof. Wolfgang Alschner's course, Data Science for Lawyers (https://www.datascienceforlawyers.org/learning-resources/lesson-8/).
This lesson uses machine learning to make predictions regarding a judge's voting record. It involves a dataset that contains Justice Brennan's SCOTUS voting record from the 1950s-1980s.
This project uses a deep learning model adapted from the classification project in unit 4 of Codecademy's skills path, Build Deep Learning Models with TensorFlow.
Files:
- WJBrennan-Voting - 3 layer neural network, randomly splits the dataset into testing and training sets
- WJBrennan_1980s_Testing - 4 layer deep neural network, splits dataset so that 1980s data is used for testing
- WJBrennan_voting - CSV dataset from Prof. Alschner's course
The first model randmonly splits the dataset (test vs. training). Predictive accuracy seems quite good at ~78%.
The second model splits the dataset so that the 1980s voting record is used for testing. Model accuracy drops significantly. With tuning, the best result I've achieved so far is 69% (but sometimes the result is much poorer).