Integrating ROC and sklearn for educational insights

• Day: 2025-06-17
• Time: 06:00 to 07:00
• Project: Teaching
• Workspace: WP 1: Strategic / Growth & Development
• Status: Completed
• Priority: MEDIUM
• Assignee: Matías Nehuen Iglesias
• Tags: ROC, Sklearn, Education, Machine Learning, AUC, Decision Trees

Description

Session Goal

The session aimed to explore the integration of ROC curve visualization with the technical content of roc_auc_score, roc_curve, and RocCurveDisplay from sklearn to develop assessments that promote deep learning among students.

Key Activities

• Detailed exploration of the RocCurveDisplay.from_predictions() method from sklearn.metrics, highlighting its educational utility and key parameters for teaching the ROC curve.
• Explanation of predict_proba() in decision trees and logistic regression, detailing probability calculations and implications for evaluation pipelines.
• Discussion on decision trees’ implicit 50% threshold for converting probabilities into binary decisions and its differences from logistic regression.
• Enhancement of statements related to AUC and threshold analysis, focusing on technical clarity and pedagogical maturity.
• Explanation of AUC’s independence from specific thresholds in classification models.
• Presentation of pedagogical strategies to encourage reflection on the trade-offs between precision, sensitivity, F1-score, and decision thresholds in classification models.
• Analysis of claims about the use of machine learning models in sklearn, including overfitting and variable scaling.
• Discussion of decision trees’ limitations in handling categorical variables, emphasizing the need for encoding and its impact on model quality.

Achievements

• Developed a comprehensive understanding of ROC curve integration in educational contexts.
• Clarified the use of predict_proba() and decision thresholds in machine learning models.
• Improved educational statements and strategies for teaching classification metrics.

Pending Tasks

• Further exploration of pedagogical frameworks for teaching machine learning metrics.
• Development of additional educational templates and guides to enhance student understanding of model evaluation techniques.

Evidence

• source_file=2025-06-17.sessions.jsonl, line_number=2, event_count=0, session_id=88c7f5be99314cfcfa7ec6a8dbd4177f6cbc69f4c5b3d8dc77dd5529abb7b752
• event_ids: []