This document summarizes a project to apply machine learning models to predict outcomes of cases in the US Court of Appeals based on historical data. It describes the data source and characteristics, data preprocessing steps including handling missing values and complex variables, dimensionality reduction techniques, models selected like random forest, neural networks and XGBoost, and the results of model tuning and testing. The neural network achieved an accuracy of 85% on test data after oversampling the training data, while XGBoost achieved 91% accuracy after tuning. The most significant variables identified for prediction included the circuit court, verdict of the previous court, nature of the appellant, and directionality of the third judge.