DS 101: Machine Learning & Modelling in Theory & Practice

Modeling is a fundamental aspect of any science. This fact is particularly apparent in data science. The key aspects of modeling that make it important for science are: (1) models are representations of things that cannot be fully understood or known (e.g., predictive models are essential to predict a future outcome, unless you have access to a time machine that we don't know about); (2) models can give us new insights into those things, including their behaviors, responses, and characteristics (especially in previously unseen conditions), thereby potentially revealing causal factors for observed outcomes and informing prescriptive actions to optimize outcomes; (3) models provide testable predictions to validate our assumptions and hypotheses about things (otherwise, it's not science); and (4) models can help answer questions that are not otherwise answerable (e.g., we can pose "what if" scenarios safely in a model environment that we would not be able or allowed to test in a real life situation). In data science, we use observation (data, evidence) to inform and inspire our models, we use machine learning (algorithms that learn from patterns in the data) to build testable models, and we use the scientific method to verify, validate, and/or refine our models. The ideal goal of these activities is discovery from data, specifically actionable insights discovery.