A Swimming Athlete Performance Prediction Model Utilizing Design Thinking and the Decision Tree Approach

Received: 15 November 2025 | Revised: 5 December 2025 | Accepted: 22 December 2025 | Online: 9 January 2026

Corresponding author: Anugerah Widi

Abstract

Traditional sports analytics faces an implementation gap where technically accurate machine learning models fail to achieve practical adoption due to misalignment with coaching needs. This study introduces Algorithmic-Enhanced Design Thinking (AEDT), a novel framework that embeds the C4.5 decision tree algorithm as an active ideation partner within Design Thinking's creative phase, transforming the sequential "design-then-analyze" paradigm into a concurrent "design-while-analyzing" process. This study analyzed 442 performance records from 94 youth swimmers aged 6-15 years registered with the Indonesian Swimming Federation through five iterative human-algorithm collaboration cycles. The AEDT process expanded the features from 6 to 18 variables, achieving 66.7% Novel Variable Discovery Rate, where two-thirds emerged from algorithmic pattern discovery. The model achieved 58.5% classification accuracy, significantly above the 33.3% random baseline, with a 0.90 interpretability score, yielding the highest Practical Value Score of 0.776 compared to Standard C4.5 at 0.687, Random Forest at 0.333, and Neural Networks at 0.234. Key discoveries include critical performance thresholds such as Height greater than 145 cm and Armspan-to-Height ratio greater than 1.02, along with stroke-specific regression models achieving R² values up to 0.882 for Freestyle prediction. The algorithm revealed that height, with an information gain of 0.892, and armspan, with 0.847, dominate performance prediction, challenging coaching assumptions about weight importance at 0.412. AEDT successfully bridges the implementation gap by producing interpretable, actionable insights through human-algorithm collaboration. The framework provides a reproducible methodology for human-AI collaboration, with collaboration quality assessed through four measurable components, namely, Human contribution H(t), Algorithm discovery A(t), Integration quality I(H, A, t), and Convergence efficiency C(t), applicable across domains where understanding "why" matters more than optimizing accuracy alone.