Pengembangan Quantum Kernel pada Quantum Machine Learning untuk Klasifikasi Grade Teh Hijau

Widigdo, Muhammad Gesang Ridho (2026) Pengembangan Quantum Kernel pada Quantum Machine Learning untuk Klasifikasi Grade Teh Hijau. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025221216-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (9MB) | Request a copy

Abstract

Penilaian kualitas Teh Hijau secara tradisional masih mengandalkan metode organoleptik yang mahal dan bersifat subjektif. Penggunaan Electronic nose (E-nose) memberikan solusi yang objektif, namun intensitas aroma yang rendah pada sampel teh kering menyebabkan pola data antarkelas saling tumpang tindih, sehingga algoritma machine learning klasik seperti Support Vector Machine (SVM) kesulitan menarik batas kesimpulan yang jelas. Quantum Machine Learning untuk klasifikasi multikelas pada grade teh hijau dengan lima kelas (A–E). Pada tahap pre-process diterapkan StratifiedGroupKFold berdasarkan Sampling_ID, standardisasi dan reduksi dimensi PCA. Komputasi kernel menggunakan pendekatan Projected Quantum Kernel (PQK). Penelitian membandingkan peta fitur IQP dengan peta fitur usulan Single-Centered Qubit (SCQ) berbasis Star-Topology Register pada model QSVC, QXGB, dan QCAT, lalu dikomparasi terhadap model klasik seperti SVM, XGBoost, CatBoost serta deep learning berupa MLP dan 1D-CNN menggunakan composite score. Hasil menunjukkan peta fitur SCQ unggul atas seluruh topologi IQP di ketiga model, dengan QSVC-SCQ sebagai model terbaik (composite 0,9458). Peningkatan paling signifikan terlihat pada kelas minoritas B (F1 naik dari 0,669–0,770 menjadi 0,922–0,960) serta generalisasi yang lebih baik (selisih F1 latih- uji 0,0866–0,0938). Pada komparasi best vs best, QSVC-SCQ mengungguli model klasik terbaik di hampir seluruh metrik.
=====================================================================================================================================
Green tea quality assessment traditionally still relies on organoleptic methods that are costly and subjective. The use of an Electronic Nose (E-nose) provides an objective solution; however, the low aroma intensity of dry tea samples causes inter-class data patterns to overlap, making it difficult for classical machine learning algorithms such as Support Vector Machine (SVM) to draw clear decision boundaries. This research develops Quantum Machine Learning for multiclass classification of green tea grades across five classes (A–E). In the pre-processing stage, StratifiedGroupKFold based on Sampling_ID is applied, along with standardization and PCA dimensionality reduction. Kernel computation adopts the Projected Quantum Kernel (PQK) approach. The research compares the IQP feature map with the proposed Single- Centered Qubit (SCQ) feature map based on a Star-Topology Register across the QSVC, QXGB, and QCAT models, which are then benchmarked against classical models such as SVM, XGBoost, and CatBoost as well as deep learning models in the form of MLP and 1D-CNN using a composite score. The results show that the SCQ feature map outperforms all IQP topologies across the three models, with QSVC-SCQ as the best model (composite 0.9458). The most significant improvement is observed in the minority class B (F1 rising from 0.669– 0.770 to 0.922–0.960) along with better generalization (train-test F1 gap of 0.0866–0.0938). In a best-vs-best comparison, QSVC-SCQ surpasses the best classical model across nearly all metrics.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Instantaneous Quantum Polynomial, Pauli, Quantum Kernel Estimation, Quantum Machine Learning, Quantum Support Vector Machine
Subjects:	Q Science > QA Mathematics > QA336 Artificial Intelligence
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Muhammad Gesang Ridho Widigdo
Date Deposited:	28 Jul 2026 01:12
Last Modified:	28 Jul 2026 01:12
URI:	http://repository.its.ac.id/id/eprint/137528

Actions (login required)

View Item