On Building Design Guidelines for An Interactive Machine Learning Sandbox Application

There are several machine learning suites that are readily-available. However, these applications require a basic foundation in machine learning making them appear difficult to configure. We introduce a sandbox approach with the goal of designing alternative programming interactions for machine learning tasks. A set of guidelines have been drafted and supported with user interviews to validate the proposed design framework. Ten students with novice machine learning experience participated in the study to formulate a programming pipeline that was used to draft the guidelines based on the literary review. A problem statement was formed from the analysis of interview insights using UX Research techniques. The insights suggest that a visual sandbox approach helps reduce the learning curve of programming machine learning tasks. The design guidelines we drafted focused on the three design factors namely system intent, interaction, and algorithm visualization. Considering these guidelines, a prototype was produced that will undergo future testing and validation.