This is my program for the Kaggle Epilepsy Seizure Prediction challenge, in which the goal was to perform machine learning on temporal EEG data to predict the onset of seizure. It is structured to allow rapid prototyping of feature selection for machine learning by creating a framework to recursively define new features based off of prior ones, uses a dependency graph to efficiently preprocess the data, and caches the results of the preprocessing step in SQLite. A similar framework is also used to define tasks that involve processing the data to allow parts of the program to be run at a time. Although my model only scored about 55% in terms of prediction accuracy, the classifier that gave the best results for me was SVM with a gaussian kernel. The main difficulty that I had from this project was difficulty cross-validating my results due to the large imbalance between the number of sample clips from patients in the interictal and preictal phases. In addition, the enormous dataset and the high number of numerical operations needed to preprocess it made training the classifier model on the whole dataset extremely time consuming. This reduced the number of training samples that I was able to use.