Statistical and Machine Learning for Inference, Prediction and Reasoning

Catalog description

In this course, we will use an open data science tool chain to develop reproducible data analyses useful for statistical and machine learning modeling for inference, prediction and reasoning on the behavior of complex systems. In addition to the standard data cleaning, assembly and exploratory data analysis steps essential to all data analyses, we will identify statistically significant relationships from datasets derived from population samples, and infer the reliability of these findings. We will use regression methods to model a number of both real-world and lab-based systems producing predictive models applicable in comparable populations. We will assemble and explore real-world datasets, perform clustering, self-similarity, and dimension reduction and linear and logistic regression to develop both fixed-effect and mixed-effect predictive models. We will introduce machine-learning approaches for classification and tree-based methods. We will use deep learning methods such as TensorFlow and PyTorch to develop neural network models of complex systems. Results will be interpreted, visualized and discussed. We will introduce the basic elements of data science and analytics using R Project open source software. R is an open-source software project with broad abilities to access machine-readable open-data resources, data cleaning and assembly functions, and a rich selection of statistical and deep learning packages, used for data analytics, model development, inference prediction and reasoning. With this background, it becomes possible to train linear regression, structural equation, fixed-effects and mixed-effects models along with other machine and deep learning models, while exploring statistically significant relationships. The class will be structured to have a balance of theory and practice. We split class sessions into Foundation and Practicum a) Foundation: lectures, presentations, discussion b) Practicum: coding, demonstrations and hands-on data science work. The M section of DSCI 353 is for students focusing on Materials Data Science. Offered as DSCI 353 , DSCI 353M and DSCI 453 . Prereq: DSCI 351 or DSCI 351M .