Advancing privacy and security in machine learning through homomorphic encryption and explainable AI

dc.authorid0000-0002-8637-5473
dc.contributor.advisorÇeliktaş, Barışen_US
dc.contributor.authorAbou Harb, Mhd Rajaen_US
dc.contributor.otherIşık Üniversitesi, Lisansüstü Eğitim Enstitüsü, Bilgisayar Mühendisliği Doktora Programıen_US
dc.contributor.otherIşık University, School of Graduate Studies, Ph.D. in Computer Engineeringen_US
dc.date.accessioned2026-04-22T12:19:24Z
dc.date.available2026-04-22T12:19:24Z
dc.date.issued2026-03-05
dc.departmentIşık Üniversitesi, Lisansüstü Eğitim Enstitüsü, Bilgisayar Mühendisliği Doktora Programıen_US
dc.departmentIşık University, School of Graduate Studies, Ph.D. in Computer Engineeringen_US
dc.descriptionText in English ; Abstract: English and Turkishen_US
dc.descriptionIncludes bibliographical references (leaves 111-124)en_US
dc.descriptionxiii, 129 leavesen_US
dc.description.abstractThe importance of data privacy in cloud-based Machine Learning is paramount, particularly in sectors such as healthcare and finance. Balancing robust privacy protection with high model accuracy remains a significant challenge. In this study, we propose a privacy-preserving framework utilizing ANNs on homomorphically encrypted data. To mitigate the computational complexity of non-linear activation functions (Sigmoid and Tanh), we developed lightweight, ANN-based estimators specifically designed for encrypted environments. Our experimental results demonstrate that these estimators significantly outperform traditional polynomial and piecewise linear methods, reducing MSE by up to 96% while improving accuracy and F1-scores. Our method achieved 97.70% accuracy and 0.9997 AUC on the MNIST dataset, validating its effectiveness. In real-world applications, we applied the approach to dyslexia detection using QEEG data, observing only minor performance degradation (2.66% accuracy, 3.86% AUC) compared to plaintext inference. Furthermore, a case study on the UCI Heart Disease dataset yielded 85.25% accuracy in encrypted inference, matching plaintext performance. Finally, we integrated the SHAP algorithm to ensure transparency for encrypted outputs. Our findings confirm that this approach successfully balances privacy, performance, and explainability, making it highly suitable for sensitive ML applications.en_US
dc.description.abstractBulut tabanlı makine öğrenimi çözümlerinde, özellikle sağlık ve finans gibi hassas alanlarda veri gizliliği kritik bir öneme sahiptir. Gizlilik koruması ile yüksek model başarımı arasında denge kurmak ise güncel bir zorluktur. Bu çalışmada, homomorfik şifreleme yöntemleriyle korunan veriler üzerinde çalışan, gizlilik odaklı bir Yapay Sinir Ağı (YSA) yaklaşımı öneriyoruz. Şifreli ortamlarda hesaplama maliyeti yüksek olan Sigmoid ve Tanh gibi doğrusal olmayan aktivasyon fonksiyonlarını verimli yönetmek amacıyla, hafif YSA tabanlı tahminciler geliştirdik. Elde edilen sonuçlar, önerdiğimiz tahmincilerin geleneksel polinom ve parçalı doğrusal yöntemlere göre üstün olduğunu; Ortalama Karesel Hatayı (MSE) %96 oranında azalttığını göstermektedir. MNIST veri kümesinde elde edilen %97,70 doğruluk ve 0,9997 AUC değerleri, yöntemin etkinliğini kanıtlamıştır. Gerçek dünya senaryolarında, QEEG verileriyle disleksi tespitinde düz metin çıkarımına kıyasla ihmal edilebilir performans kayıplarıyla (%2,66 doğruluk, %3,86 AUC) başarı sağlanmıştır. UCI Kalp Hastalığı veri kümesinde ise düz metin performansıyla eşdeğer %85,25 doğruluk elde edilmiştir. Ayrıca, şeffaflığı artırmak amacıyla şifreli çıkarımlara SHAP tabanlı açıklanabilirlik entegre edilmiştir. Bulgularımız, önerilen modelin gizlilik, yüksek performans ve açıklanabilirlik gereksinimlerini başarıyla dengelediğini ve hassas sektörler için güçlü bir çözüm sunduğunu ortaya koymaktadır.en_US
dc.description.tableofcontentsCLOUD-BASED MACHINE LEARNINGen_US
dc.description.tableofcontentsVERVIEW OF PPMLen_US
dc.description.tableofcontentsCHALLENGES OF NON-LINEARITY IN HOMOMORPHICALLY ENCRYPTED ML MODELSen_US
dc.description.tableofcontentsRIVACY-PRESERVING AND EXPLAINABLE AI IN HEALTHCARE: INSIGHTS FROM DYSLEXIA DETECTIONen_US
dc.description.tableofcontentsESEARCH GAPS IN PRIVACY-PRESERVING AND EXPLAINABLE MACHINE LEARNINGen_US
dc.description.tableofcontentsANNsen_US
dc.description.tableofcontentsMathematical Foundation of ANNsen_US
dc.description.tableofcontentsLearning Process of ANNsen_US
dc.description.tableofcontentsActivation Functions in ANNsen_US
dc.description.tableofcontentsHEen_US
dc.description.tableofcontentsMathematical Foundation of CKKSen_US
dc.description.tableofcontentsXAIen_US
dc.description.tableofcontentsXAI Typesen_US
dc.description.tableofcontentsSHAPen_US
dc.description.tableofcontentsCHALLENGES WITH NON-LINEAR FUNCTIONS IN HEen_US
dc.description.tableofcontentsPREVIOUS SOLUTIONS FOR NONLINEARITY IN HOMOMORPHICALLY ENCRYPTED ANNen_US
dc.description.tableofcontentsPolynomial Approximationen_US
dc.description.tableofcontentsPiecewise Linear Approximationen_US
dc.description.tableofcontentsREGULATORY FRAMEWORKS AND DATA PRIVACYen_US
dc.description.tableofcontentsGDPRen_US
dc.description.tableofcontentsHIPAAen_US
dc.description.tableofcontentsKVKKen_US
dc.description.tableofcontentsEnsuring Regulatory Compliance with Homomorphic Encryption and Explainable AIen_US
dc.description.tableofcontentsMOTIVATION FOR ANN-BASED ESTIMATORSen_US
dc.description.tableofcontentsOVERVIEW OF THE PROPOSED SOLUTIONen_US
dc.description.tableofcontentsDESIGN OF THE MAIN ANNen_US
dc.description.tableofcontentsMain ANN for MNIST Classificationen_US
dc.description.tableofcontentsMain ANN for Dyslexia Detectionen_US
dc.description.tableofcontentsDESIGNS OF THE ANN ESTIMATORSen_US
dc.description.tableofcontentsHOMOMORPHICALLY ENCRYPTED INFERENCEen_US
dc.description.tableofcontentsEXPLAINABILITY IN PPMLen_US
dc.description.tableofcontentsMNIST Dataseten_US
dc.description.tableofcontentsQEEG Dataset for dyslexia detectionen_US
dc.description.tableofcontentsHARDWARE AND SOFTWARE ENVIRONMENTen_US
dc.description.tableofcontentsHOMOMORPHIC ENCRYPTION SCHEME SETTINGSen_US
dc.description.tableofcontentsCASE STUDY: PREDICTING HEART DISEASEen_US
dc.description.tableofcontentsEVALUATION MEASUREMENTen_US
dc.description.tableofcontentsMetrics and Procedures for Estimator Assessmenten_US
dc.description.tableofcontentsEvaluation Metrics for Classification Modelsen_US
dc.description.tableofcontentsEvaluation Metrics for Explainable AI under Homomorphic Encryptionen_US
dc.description.tableofcontentsERFORMANCE OF STANDALONE ACTIVATION FUNCTION ESTIMATORSen_US
dc.description.tableofcontentsHOMOMORPHIC ENCRYPTION INFERENCE ON THE MNIST DATASETen_US
dc.description.tableofcontentsResults Using Sigmoid Activation Function Estimatorsen_US
dc.description.tableofcontentsResults Using Tanh Activation Function Estimatorsen_US
dc.description.tableofcontentsEncrypted Inference Timeen_US
dc.description.tableofcontentsComparative Analysisen_US
dc.description.tableofcontentsREAL-WORLD APPLICATION: HE INFERENCE AND EXPLAINABLE AI FOR DYSLEXIA DETECTIONen_US
dc.description.tableofcontentsCASE STUDY RESULTS: HEART DISEASE PREDICTIONen_US
dc.description.tableofcontentsSECURITY THREAT MODEL AND GAME-THEORETIC ANALYSISen_US
dc.description.tableofcontentsGame-Theoretic Security Frameworken_US
dc.description.tableofcontentsMitigation Strategiesen_US
dc.description.tableofcontentsANALYSIS OF STANDALONE ACTIVATION FUNCTION ESTIMATORSen_US
dc.description.tableofcontentsEVALUATING HOMOMORPHIC ENCRYPTION INFERENCE ON THE MNIST DATASETen_US
dc.description.tableofcontentsNSIGHTS FROM HE INFERENCE AND EXPLAINABLE AI IN DYSLEXIA DETECTION USING QEEG DATAen_US
dc.description.tableofcontentsDyslexia Classification Discussionen_US
dc.description.tableofcontentsAnalysis of SHAP Perturbation Sensitivity in Encrypted Inferenceen_US
dc.description.tableofcontentsSHAP Analysis and Neurophysiological Interpretabilityen_US
dc.description.tableofcontentsImplications for Research and Clinical Applicationsen_US
dc.description.tableofcontentsCASE STUDY ANALYSIS: UCI HEART DISEASE DATASETen_US
dc.identifier.citationAbou Harb, M. R. (2026). Advancing privacy and security in machine learning through homomorphic encryption and explainable AI. İstanbul: Işık Üniversitesi Lisansüstü Eğitim Enstitüsü.en_US
dc.identifier.urihttps://hdl.handle.net/11729/7336
dc.institutionauthorAbou Harb, Mhd Rajaen_US
dc.institutionauthorid0000-0002-8637-5473
dc.language.isoenen_US
dc.publisherIşık Üniversitesi, Lisansüstü Eğitim Enstitüsüen_US
dc.relation.publicationcategoryTezen_US
dc.rightsinfo:eu-repo/semantics/openAccessen_US
dc.subjectPrivacy-preserving machine learningen_US
dc.subjectHomomorphic encryptionen_US
dc.subjectArtificial neural networksen_US
dc.subjectExplainable AIen_US
dc.subjectSHAPen_US
dc.subjectEncrypted inferenceen_US
dc.subjectGizlilik koruyucu makine öğrenimien_US
dc.subjectHomomorfik şifrelemeen_US
dc.subjectYapay sinir ağlarıen_US
dc.subjectAçıklanabilir yapay zekâen_US
dc.subjectŞifreli çıkarımen_US
dc.titleAdvancing privacy and security in machine learning through homomorphic encryption and explainable AIen_US
dc.title.alternativeMakine öğreniminde gizliliğin ve güvenliğin ilerletilmesi: homomorfik şifreleme ve açıklanabilir yapay zekâ üzerine bir çalışmaen_US
dc.typeDoctoral Thesisen_US
dspace.entity.typePublicationen_US

Dosyalar

Orijinal paket
Listeleniyor 1 - 1 / 1
Yükleniyor...
Küçük Resim
İsim:
Advancing_privacy_and_security_in_machine_learning_through_homomorphic_encryption_and_explainable_AI.pdf
Boyut:
2.57 MB
Biçim:
Adobe Portable Document Format
Lisans paketi
Listeleniyor 1 - 1 / 1
Küçük Resim Yok
İsim:
license.txt
Boyut:
1.17 KB
Biçim:
Item-specific license agreed upon to submission
Açıklama: