Optimasi Prediksi Kemunculan Bug Pada Perangkat Lunak Dalam Pengembangan Agile Menggunakan Algoritma Seleksi Fitur

Wicaksono, Faiq Dhimas (2025) Optimasi Prediksi Kemunculan Bug Pada Perangkat Lunak Dalam Pengembangan Agile Menggunakan Algoritma Seleksi Fitur. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6032231111-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (6MB) | Request a copy

Abstract

Dalam pengembangan perangkat lunak berbasis Agile, kebutuhan akan deteksi cacat secara dini menjadi semakin penting seiring meningkatnya frekuensi rilis dan perubahan kebutuhan yang cepat. Metode konvensional yang hanya bergantung pada analisis kode atau commit log sering kali gagal menangkap sinyal kontekstual penting yang bersumber dari aktivitas pengembangan modern. Tantangan ini diperparah oleh karakteristik data Agile yang dinamis, tidak terstruktur, dan berumur pendek, sehingga menuntut model prediksi yang adaptif dan efisien terhadap perubahan.
Penelitian ini bertujuan untuk mengembangkan dan mengevaluasi sebuah kerangka kerja pembelajaran mesin hibrida yang mampu memprediksi keberadaan bug secara efektif menggunakan data issue ticket dari Jira. Kerangka kerja ini mengombinasikan fitur-fitur kontekstual dari 981 tiket pengembangan perangkat lunak Tokopedia (Maret 2022) dengan teknik seleksi fitur berbasis algoritma metaheuristik, yaitu Bat Algorithm (BA) dan Particle Swarm Optimization (PSO). Berbagai model klasifikasi diterapkan, meliputi Random Forest (RF), Support Vector Machine (SVM), XGBoost, CatBoost, dan Transformer, guna membandingkan performa prediksi cacat pada data yang kompleks dan heterogen. Evaluasi dilakukan menggunakan metrik menggunakan sejumlah metrik performa, termasuk precision, recall, accuracy, F1-score, dan AUC-ROC, dengan fokus utama pada F1-score dan AUC-ROC untuk mencerminkan keseimbangan antara sensitivitas dan ketahanan klasifikasi. Solusi ini diharapkan mampu membantu tim pengembang mengidentifikasi risiko bug secara lebih tepat dan cepat, serta mendukung keputusan perbaikan dalam siklus sprint yang singkat.
Hasil eksperimen menunjukkan bahwa RF yang dioptimasi dengan BA memberikan performa terbaik, dengan F1-score 0,83 dan AUC-ROC 0,86. Model boosting juga memberikan hasil kompetitif, namun kurang efisien secara komputasi. Di sisi lain, SVM dan Transformer unggul dalam recall, namun memiliki presisi dan AUC yang rendah. Secara keseluruhan, pendekatan RF dengan optimasi BA terbukti efektif dalam meningkatkan kualitas perangkat lunak dan mendukung manajemen bug yang adaptif dalam lingkungan Agile.
=======================================================================================================================================
In Agile-based software development, the need for early bug detection is becoming increasingly crucial as release cycles accelerate and requirements shift rapidly. Conventional approaches that rely solely on code analysis or commit logs often fall short in capturing important contextual signals generated during modern development activities. This challenge is further intensified by the nature of Agile data, which is dynamic, unstructured, and short-lived, necessitating predictive models that are both adaptive and efficient in response to change.
This study aims to develop and evaluate a hybrid machine learning framework capable of effectively predicting bug presence using issue ticket data from Jira. The proposed framework combines contextual features extracted from 981 software development tickets collected from Tokopedia (March 2022) with feature selection techniques based on metaheuristic algorithms, namely Bat Algorithm (BA) and Particle Swarm Optimization (PSO). Various classification models are applied, including Random Forest (RF), Support Vector Machine (SVM), XGBoost, CatBoost, and Transformer, to compare predictive performance on complex and heterogeneous data. The evaluation uses multiple performance metrics, including precision, recall, accuracy, F1-score, and AUC-ROC, with F1-score and AUC-ROC emphasized as the primary indicators to balance sensitivity and classification robustness. This solution is expected to assist development teams in identifying bug risks more accurately and efficiently, thereby supporting timely corrective decisions within Agile sprint cycles.
Experimental results show that the RF model optimized with BA achieves the best performance, with an F1-score of 0,83 and AUC-ROC of 0,86. Boosting models also produce competitive results but are less computationally efficient. Meanwhile, SVM and Transformer excel in recall but suffer from lower precision and AUC. Overall, the RF-BA approach proves to be effective in enhancing software quality and supporting adaptive bug management in Agile environments.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Prediksi Cacat, Pengembangan Agile, Seleksi Fitur, Algoritma Optimasi, Kualitas Perangkat Lunak, Bug Prediction, Agile Development, Feature Selection, Optimization Algorithms, Software Quality
Subjects:	T Technology > T Technology (General) > T56.8 Project Management T Technology > T Technology (General) > T57.5 Data Processing T Technology > T Technology (General) > T57.8 Nonlinear programming. Support vector machine. Wavelets. Hidden Markov models. T Technology > T Technology (General) > T58.6 Management information systems T Technology > T Technology (General) > T58.62 Decision support systems T Technology > T Technology (General) > T58.8 Productivity. Efficiency
Divisions:	Interdisciplinary School of Management and Technology (SIMT) > 61101-Master of Technology Management (MMT)
Depositing User:	Faiq Dhimas Wicaksono
Date Deposited:	31 Jul 2025 02:03
Last Modified:	31 Jul 2025 02:03
URI:	http://repository.its.ac.id/id/eprint/124267

Actions (login required)

View Item