CS040 Data Science and Statistics Education

The abstract is optional for extended abstract so we do not have a short abstract. 

Data science education requires students to integrate statistical and computational thinking. There is a recognized need for effective and scalable methods for teaching computer programming to introductory data science students, particularly given the limited research on the pedagogical design of interactive coding tasks. We explain our approach to designing interactive worked examples, which utilize computational tools to incorporate interactivity, such as editable code executed in a browser and auto-marked questions. Our design is based on epistemic programming, viewing programming as intertwined with knowledge acquisition through a tinkering approach where students iteratively adjust and reflect on code changes. Essential to this design are programming plans, integrated with "code blanks" that guide adaptation, following a "fading out" technique. We illustrate our design approach using an interactive worked example from an introductory R course that uses data sourced from the NASA POWER API. This interactive worked example employed sequenced programming plans and "tinker questions" to prompt students to modify code blanks, practice syntax, and reflect on data outcomes. To guide the development of interactive worked examples for introductory-level data science students, we propose four essential design principles: presenting a carefully sequenced collection of programming plans; utilising programming plans with code blanks; employing tinker questions to guide code completion and subsequent reflection; and letting students engage in an individual project after engaging with the worked example. We argue that learning programming should be framed epistemically, fostering curiosity and a tinkering mindset. We propose that well-designed interactive worked examples are essential for delivering the conceptual and pedagogical support required for data science programming, a role automated Generative AI tools cannot easily replicate alone. 

This talk introduces a mobile app to teach, learn and apply Machine Learning, designed to make core algorithms accessible to students and instructors across backgrounds, including those without coding experience. In line with GAISE (2016) recommendations promoting conceptual understanding and hands-on data analysis, the module leverages smartphones and tablets as an immediately available, consistent, and low-barrier technological platform. The app provides interactive implementations of classical ML methods such as Naïve Bayes, Linear Discriminant Analysis, and K-Means Clustering along with built-in datasets or user-uploaded CSV files. Through demonstrations using the Palmer Penguins dataset, the presentation will highlight how the module supports key ML concepts including feature selection, train–test splits, visualization and interpretation of posterior odds, classification boundaries, model accuracy, and prediction for new observations. Attendees will see how the app can serve as the primary technological tool for teaching ML in courses that emphasize interpretation and understanding rather than (or along with) programming. All participants will be able to download the app during the session and will leave with practical strategies for integrating mobile-based ML instruction into their courses.