Smart Diagnosis of Cervical Cancer Through Comparative Study of Machine Learning Algorithms for Early Prediction Melisa Krisnawati Fulfillment of the Requirement for the Degree of = 자궁경부암의 조기 예측을 위한 머신러닝 알고리즘 비교 연구를 통한 스마트 진단 시스템

다국어 초록 (Multilingual Abstract)

Cervical cancer is one of the most critical health concernsamong women worldwide, particularly in low and middle-incomecountries where early screening and medical resources remain limited. Despite advances in medical imaging and molecular diagnostics, conventional screening methods such as Pap smears and HPVtestingoften require specialized equipment and are subject to humaninterpretation errors, leading to delayed or inaccurate diagnosis. Toaddress these challenges, this study proposes a smart diagnosticframework that utilizes machine learning and deep learning techniquesto enhance early prediction and risk assessment of cervical cancer. Using the publicly available cervical cancer dataset fromtheUCI repository, the research conducts a comparative study among several algorithms, including Random Forest (RF), Decision Tree (DT), Logistic Regression (LR), and deep learning models such as Multilayer
Perceptron (MLP) and Convolutional Neural Network (CNN). Toensure data quality and robustness, data preprocessing techniques wereapplied, including Recursive Feature Elimination (RFE) for featureselection and Synthetic Minority Oversampling Technique (SMOTE) toaddress class imbalance. These approaches help the model focus onthemost relevant risk factors while preventing bias caused by unequal datadistribution. The results from the comparative analysis indicate that
integrating feature selection and class balancing significantly improvespredictive reliability and model stability. Furthermore, the inclusionof
deep learning within the ensemble framework enhances the adaptabilityof the model when dealing with complex and nonlinear patterns withinthe dataset. The proposed system demonstrates potential as a decision- support tool that assists medical practitioners in early screeningandprevention efforts, providing a foundation for data-driven healthcaresolutions and supporting global initiatives to reduce cervical cancer
mortality.

국문 초록 (Abstract)

자궁경부암은 전 세계 여성에게가장심각한건강위협중하나로, 특히 의료 자원과 조기 검진이 부족한개발도상국에서그발생률과 사망률이 높다. 기존의 자궁경부암검사방법인자궁경부 세포검사(Pap smear)나 HPV 검사 등은전문장비와숙련된인력을 필요로 하며, 검사자의 주관적 판단에따라오진이나진단지연이 발생할 가능성이 있다. 이러한 한계를극복하기위해본연구에서는 머신러닝과 딥러닝 기법을 활용한스마트진단프레임워크를제안하여 자궁경부암의 조기 예측과 위험요인분석의효율성을높이고자 한다. 본 연구는 UCI 에서 공개된 자궁경부암데이터셋을기반으로 머신러닝 알고리즘 간 비교 연구를 수행하였다. 랜덤포레스트(Random Forest), 의사결정나무(Decision Tree), 로지스틱회귀(Logistic Regression) 등의 전통적 모델과더불어다층퍼셉트론(MLP), 합성곱 신경망(CNN)과 같은 딥러닝모델을적용하였다. 데 이터 품질과 모델의 일반화 성능을 향상시키기위해재귀적특성제거(RFE) 를 활용한 주요 변수 선택과 합성소수클래스오버샘플링 기법(SMOTE)을 통한 데이터 불균형보정을수행하였다. 이를통해 모델은 중요한 위험 요인에 집중하고, 데이터편향으로인한성능 저하를 방지할 수 있었다. 비교 분석 결과, 특성 선택과 데이터균형화방법을결합하면 예측 신뢰도와 모델의 안정성이 크게향상됨을확인하였다. 또한 딥러닝 기반의 앙상블 구조는 비선형적이고복잡한데이터패턴을 학습하여 더 높은 예측 성능을 보였다. 제안된프레임워크는의료 전문가의 조기 진단 및 예방 전략 수립을지원할수있는지능형 의사결정 보조 도구로서의 가능성을보여주며, 데이터기반헬스케어 발전과 자궁경부암 사망률 감소를위한국제적노력에기여할 수 있을 것으로 기대된다

목차 (Table of Contents)

• List of Contents
• List of Contents i
• List of Tables iii
• List of Figures iv
• List of Abbreviations ⅴ
• Acknowledgement ⅵ
• 1. Introduction 1
• 1.1 Research Motivation. 2
• 1.2 Novelty and Contributions. 4
• 1.2.1 Main Contributions. 4
• 1.3 Thesis Organization 5
• 2. Background. 6
• 2.1 Logistic Regression 7
• 2.2 Support Vector Machine 7
• 2.3 K-Nearest Neighbors (KNN) 8
• 2.4 Random Forest 9
• 2.5 Naive Bayes 10
• 2.6 Convolutional Neural Network (CNN) 10
• 2.7 Multilayer Perceptron (MLP) 11
• 3. Related Work. 13
• 4. Methodology 16
• 4.1 Dataset and Preprocessing 17
• 4.2 Feature Selection and SMOTE 25
• 4.3 Model Development 26
• 4.4 Evaluation Metrics 28
• 5. Experiments Results 29
• 5.1 Algorithm Comparison. 29
• 5.2 Discussion 34
• 6. Conclusion. 36
• References. 38
• ABSTRACT... 42
• 국문요약 44