Integrating Human-Computer Interaction Principles into the Interface Design of ANALYTiC Platform to Enhance Usability

Sammanfattning

Active Learning (AL) systems aim to reduce labeling effort by iteratively querying human experts for the most informative data instances. However, their real-world effectiveness hinges not only on algorithmic performance but also on the usability of the interface through which experts interact with the system. This thesis investigates how integrating Human-Computer Interaction (HCI) principles can enhance the usability of ANALYTiC, an Active Learning (AL) platform for semantic trajectory classification, thereby improving task effectiveness and user experience for individuals with diverse technical backgrounds. Guided by Nielsen’s 10 Usability Heuristics and Shneiderman’s Eight Golden Rules, a heuristic evaluation of the original interface identified seven key usability problems, including poor visibility of workflow's instructions, manual label entry, fragmented results, and redundant visualizations. These findings informed a targeted redesign (Platform B) that preserved the underlying AL algorithm while improving the clarity, consistency, and feedback of the interface. A mixed-methods user study with six participants compared the original (Platform A) and redesigned interfaces using task-specific effectiveness ratings (inspired by SEQ), the System Usability Scale (SUS) and thematic analysis of qualitative feedback. The results show consistent improvements in all tasks: Platform B achieved a mean SUS score of 80.8 ("excellent" usability) versus 65.4 for Platform A ("average"), with the most substantial gains in label configuration (33\% → 100\% "very/extremely effective") and interpretation of results. Qualitative findings confirm that the redesign improved learnability, reduced cognitive friction, and increased user confidence. This work demonstrates that even modest interface changes grounded in theory can substantially improve the human experience in scientific AL systems, offering a replicable model for user-centered design in human-in-the-loop machine learning.