[Paper Review] Data, Trees, and Forests -- Decision Tree Learning in K-12 Education

Tilman Michaeli, Stefan Seegerer+2 more|arXiv (Cornell University)|May 10, 2023

Statistics Education and Methodologies4 citations

TL;DR

This paper presents a K-12 teaching concept that combines unplugged decision tree activities with Orange, a visual data analysis tool, to teach supervised learning and decision tree learning in a low-barrier, hands-on way. Students gain conceptual understanding and apply ML methods to real datasets, leading to improved awareness of ethical issues and practical ML limitations.

ABSTRACT

As a consequence of the increasing influence of machine learning on our lives, everyone needs competencies to understand corresponding phenomena, but also to get involved in shaping our world and making informed decisions regarding the influences on our society. Therefore, in K-12 education, students need to learn about core ideas and principles of machine learning. However, for this target group, achieving all of the aforementioned goals presents an enormous challenge. To this end, we present a teaching concept that combines a playful and accessible unplugged approach focusing on conceptual understanding with empowering students to actively apply machine learning methods and reflect their influence on society, building upon decision tree learning.

Motivation & Objective

• To address the challenge of teaching machine learning in K-12 education with no prior programming or CS knowledge required.
• To bridge conceptual understanding of supervised learning with hands-on application using real datasets.
• To empower students to explore, compare, and reflect on ML methods and their societal implications.
• To support interdisciplinary learning by enabling application of data science across subjects like science, geography, and ethics.
• To reduce the barrier to ML education by combining unplugged learning with a no-code visual tool (Orange) for model building.

Proposed method

• Students first manually construct decision trees using an unplugged activity (the 'monkey game') to understand the logic of decision tree learning.
• They then use Orange, a visual data analysis tool, to build and train decision trees on the same dataset, enabling comparison between manual and automated methods.
• The approach uses data flow diagrams instead of code, eliminating syntax errors and allowing focus on data interpretation and model evaluation.
• Students apply the method to real-world datasets, such as U.S. census data and academic performance data, to explore feature relevance and prediction quality.
• The teaching sequence includes a project phase where students design their own computational artifacts using ML, followed by discussions on fairness, transparency, and bias.
• Support materials, including lesson plans and datasets, are provided under a Creative Commons license for classroom use.

Experimental results

Research questions

• RQ1 How can decision tree learning be taught effectively in K-12 education without requiring programming knowledge?
• RQ2 To what extent does combining unplugged activities with a visual data analysis tool like Orange improve conceptual and practical understanding of supervised learning?
• RQ3 How do students perceive the ethical implications of ML when working with real datasets, such as those containing sensitive attributes like race?
• RQ4 Can students develop a meaningful understanding of model evaluation, training/testing splits, and the distinction between correlation and causation through this approach?
• RQ5 In what ways can this teaching concept support interdisciplinary learning across science, social studies, and humanities subjects?

Key findings

• Students developed a clear understanding of core concepts in supervised learning, including training and testing phases, model evaluation, and the limitations of ML predictions.
• The unplugged activity successfully made the internal logic of decision trees accessible and engaging, even without prior technical knowledge.
• Students were surprised and critically reflective when they discovered that seemingly irrelevant or sensitive features—like race or parental education—could be used in models.
• The use of Orange enabled students to experiment with hyperparameters and compare manual vs. automated tree construction, enhancing their grasp of model behavior.
• The teaching approach fostered awareness of ethical issues such as fairness, transparency, and bias in ML, particularly when working with real-world datasets.
• Students reported that the experience was eye-opening in demonstrating how easily large datasets can be analyzed and how quickly models can make predictions based on complex or questionable features.