Klasifikasi Boosted Top Jet Tagging Menggunakan Particlenet Dan XGBoost
Date
2026Jenis/Type
SkripsiSubtype
Undergraduate ThesesAuthor
WICAKSONO, ARJUNA DWI PUTRA
Puspita, R. Tony Ibnu Sumaryada Wijaya
Yani, Sitti
Metadata
Show full item recordAbstract
Penelitian ini bertujuan membandingkan kinerja dua pendekatan machine learning untuk klasifikasi boosted top jet, yaitu ParticleNet berbasis graph neural network dan XGBoost berbasis gradient boosting decision tree menggunakan dataset ATLAS Top Tagging Open Data. Model ParticleNet memanfaatkan informasi low level constituent berupa partikel penyusun jet. Sementara itu, model XGBoost menggunakan high level jet hasil feature engineering. Evaluasi performa model dilakukan menggunakan beberapa metrik klasifikasi, yaitu Area Under the Curve (AUC), signal efficiency, dan background rejection. Hasil penelitian menunjukkan bahwa ParticleNet memberi kinerja klasifikasi yang lebih unggul dengan nilai AUC dan background rejection yang lebih tinggi pada berbagai tingkat signal efficiency. Namun, model ini memerlukan sumber daya komputasi dan kompleksitas training yang lebih besar. Disisi lain, XGBoost tetap memberikan performa yang kompetitif dengan biaya komputasi yang lebih rendah serta interpretabilitas model yang lebih baik. Secara keseluruhan, ParticleNet lebih efektif untuk boosted top tagging, sedangkan XGBoost tetap menjadi alternatif yang lebih efisien dan mudah diinterpretasikan. This research aims to compare the performance of two machine learning approaches for boosted top jet classification: ParticleNet based on graph neural network and XGBoost based on gradient boosting decision trees, using the ATLAS Top Tagging Open Data. ParticleNet model utilizes low level constituent information in the form of jet constituent particles and applies dynamic graph convolution to capture complex spatial relationships among particles. whereas, XGBoost model use the high level features obtained through feature engineering. Model evaluation was conducted using several classification metrics, including Area under the Curve (AUC), signal efficiency and background rejection. The result show that ParticleNet achieves superior classification performance compared to XGBoost, with higher AUC and background rejection across various signal efficiency value. On the other hand, XGBoost still demonstrates competitive performance with lower level computational cost and better model interpretability. Overall, ParticleNet are more effective for boosted top tagging, while XGBoost remain as a computationally efficient and interpretable alternative.
Collections
- UF - Physics [1287]

