IS517: Methods of Data Science

Modern computing has opened up powerful new approaches to statistical modeling and analysis, which together form the methodological core of data science. This course surveys techniques for both predictive and descriptive learning, drawing on ideas from statistics, computer science, and artificial intelligence. A central theme is predictive modeling: fitting models to data in order to predict future outcomes. On the regression side, we study linear models — including modern approaches for working with large numbers of variables — together with smoothing techniques, additive models, and local regression. For classification, we cover logistic regression, regularization, discriminant analysis, splines, support vector machines, generalized additive models, naive Bayes, mixture models, and nearest-neighbor methods as time permits. Throughout the semester, each method is placed within the broader "data science lifecycle" — the set of practices through which scientific findings are discovered and communicated.

The course moves at a brisk pace and emphasizes hands-on computing exercises using Python and other relevant languages.

This is a group project. Ideally, every group should have two students. If this is not possible, please discuss with the instructor.

The project proposal will describe the proposed dataset(s), the original research question(s), and the proposed method of solution. This will likely be the novel application of a regression or classification technique from class. It is at most one page in length.

The final project will carry out the research in the proposal. There will be a project proposal presentation describing the research question(s), dataset(s), method, and possibly the expected results. There will be a final project presentation describing the research question(s), dataset(s), method, and comparing the expected and actual results.

Schedule is subject to revision.