Kyle Polich avatar

Dexa/Kyle Polich

Learn more
Published Mar 3, 2022

Explainable K-Means

Kyle Polich delves into the intricacies of explainable k-means clustering with expert Lucas Murtinho, revealing how decision trees enhance interpretability in unsupervised learning while balancing partition quality. They explore the "price of explainability" in clustering and uncover future research directions, emphasizing innovative approaches and trade-offs for algorithmic transparency.
Episode Highlights
Data Skeptic logo

Topics covered

Popular Clips

Episode Highlights

  • Tree Explanation

    Lucas Murtinho explains how decision trees can simplify the interpretation of k-means clustering results. By using binary decision trees, complex cluster boundaries are broken down into yes or no questions, making them easier to understand for non-experts. This approach allows individuals without a technical background to grasp the essence of clustering results, as Murtinho notes, "The idea is that even someone who's not familiar with machine learning can understand what's going on." 1 2

       

    Interpretability

    The challenge of making unsupervised algorithms like k-means interpretable is significant. Murtinho highlights the novelty of applying explainability to unsupervised learning, a field traditionally focused on supervised models. He emphasizes the importance of understanding machine learning models for ethical and technical reasons, stating, "Explainability is a crucial topic for machine learning, for it to be more accessible, more reliable." 1 3

       

    Algorithm Trade-offs

    Explaining clustering algorithms involves balancing tree depth and interpretability. Murtinho discusses the trade-offs between tree depth and the quality of explanations, noting that deeper trees may not always provide clearer insights. He explains, "If we don't worry about depth, the tree ends up being as deep as it can," which can hinder explainability. 4 2

Related Episodes