Penggerombolan Pemain Indonesian Basketball League 2022 Menggunakan K-Means dan K-Medoid dengan t-SNE Dimensionality Reduction

Abstract

Basket modern telah berkembang menjadi olahraga “position-less”. Posisi dan peran pemain tidak lagi ditentukan berdasarkan kriteria antropometri, melainkan karakteristik dan skill set yang dibuktikan oleh catatan statistik di lapangan. Perbedaan karakteristik dan skill set dapat diketahui dengan menggerombolkan pemain berdasarkan antropometri dan statistik pertandingan. Analisis gerombol tradisional, cenderung menghasilkan hasil kurang optimal jika diterapkan langsung pada data berdimensi tinggi karena adanya “curse of dimensionality”. Permasalahan tersebut dapat diatasi dengan reduksi dimensi. Penelitian ini bertujuan menggerombolkan pemain Indonesian Basketball League (IBL) 2022, berdasarkan antropometri dan statistik pertandingan menggunakan k-means dan k-medoid dengan reduksi dimensi t-SNE, serta menentukan metode penggerombolan terbaik. Metode penggerombolan terbaik ditentukan berdasarkan parameter optimal yang dievaluasi menggunakan koefisien Silhouette dan indeks S_Dbw. Berdasarkan hasil evaluasi, metode k-medoid pada t-SNE perplexity = 2 menghasilkan nilai evaluasi optimal, namun kedua hasil evaluasi menghasilkan banyak gerombol (k) optimal berbeda. Berdasarkan Silhouette, k-medoid pada t-SNE perplexity = 2 optimal pada k = 9, sedangkan berdasarkan indeks S_Dbw, k-medoid pada t-SNE perplexity = 2 optimal pada k = 10. Penentuan gerombol terbaik dilakukan secara eksploratif dengan radar chart dan Chernoff faces guna melihat perbedaan hasil optimal. Hasil penggerombolan terbaik yang diperoleh adalah k-medoid pada t-SNE perplexity = 2 dengan 10 gerombol optimal.

Modern basketball has evolved into a “position-less” sport. Player positions and roles are no longer determined by anthropometric criteria, but rather based on their characteristics and skill sets, as indicated by statistical field performance. Grouping players based on anthropometry and match statistics allows for the identification of differences in their characteristics and skill sets. Traditional cluster analysis tends to produce less-optimal results when applied directly to high dimensional data because of the "curse of dimensionality". This issue can be addressed through dimensionality reduction techniques. This study aims to cluster Indonesian Basketball League (IBL) 2022 players, based on anthropometry and match statistics using k-means and k-medoid with t-SNE dimensionality reduction, and determine the best clustering method. The best clustering method is determined by evaluating the optimal parameters using Silhouette coefficient and S_Dbw index. Based on the evaluation results, the k-medoid with t-SNE perplexity = 2 yields optimal evaluation values, however the two evaluations indicate different optimal cluster sizes (k). Based on the Silhouette coefficient, k-medoid at t-SNE perplexity = 2 is optimal at k = 9, while based on the S_Dbw index, k-medoid at t-SNE perplexity = 2 is optimal at k = 10. To compare the optimal results, the determination of the best cluster is further explored using radar charts and the Chernoff faces. The best clustering results were obtained using the k-medoid with t-SNE perplexity = 2 and k = 10.