EVALUASI EFEKTIVITAS TEKNIK PRIVACY-PRESERVING: K-ANONYMITY, L-DIVERSITY, T-CLOSENESS PADA DATA SENSITIF

EVALUATION OF THE EFFECTIVENESS OF PRIVACY-PRESERVING TECHNIQUES—K-ANONYMITY, L-DIVERSITY, AND T-CLOSENESS—ON SENSITIVE DATA

Abstract

Perlindungan privasi menjadi aspek krusial dalam pengumpulan, pengolahan, dan publikasi data sensitif, namun potensi risiko kebocoran informasi dapat menimbulkan konsekuensi hukum maupun kerugian reputasi. Untuk menjaga keseimbangan antara kegunaan data dan privasi individu, teknik anonimisasi menjadi pendekatan utama, termasuk penerapan k-anonymity dan evaluasi menggunakan l-diversity dan t-closeness. Penelitian ini bertujuan untuk mengevaluasi efektivitas teknik-teknik tersebut dalam mengurangi risiko pengungkapan identitas dan atribut sensitif pada dataset kesehatan. Studi kasus menggunakan 55500 dataset medis dengan quasi-identifier Age, Gender, dan Blood Type, serta atribut sensitif Medical Condition. Dataset dianonimkan menggunakan k-anonymity melalui proses generalisasi dan supresi untuk membentuk equivalence class dengan ukuran minimum k ? 5. Selanjutnya, dataset dievaluasi menggunakan l-diversity untuk mengukur keberagaman atribut sensitif dalam setiap kelompok, serta t-closeness untuk menilai kesamaan distribusi atribut sensitif terhadap distribusi global menggunakan Earth Mover’s Distance (EMD). Hasil pengujian menunjukkan bahwa seluruh equivalence class telah memenuhi k ? 5 dengan suppression rate sebesar 1,15%. Evaluasi l-diversity menunjukkan tidak terdapat equivalence class dengan l < 2, sehingga risiko attribute disclosure dapat diminimalkan. Pengujian t-closeness menggunakan Earth Mover’s Distance (EMD) menunjukkan mayoritas kelas memiliki EMD ? 0,15 dan hanya satu kelas dengan nilai sedikit di atas ambang batas t = 0,2. Dari sisi utilitas data, nilai Normalized Generalized Information Loss (NGIL) sebesar 0,079 (7,9%) dan AECS sebesar 6,28 menunjukkan tingkat kehilangan informasi yang rendah tanpa terjadi over-generalization. Secara keseluruhan, kombinasi metode yang diterapkan berhasil mencapai keseimbangan antara perlindungan privasi dan data utility.

Privacy protection has become a crucial aspect in the collection, processing, and publication of sensitive data, as potential risks of information leakage may lead to legal consequences and reputational damage. To maintain a balance between data utility and individual privacy, anonymization techniques serve as a primary approach, including the implementation of k-anonymity and its evaluation using l-diversity and t-closeness. This study aims to evaluate the effectiveness of these techniques in reducing the risk of identity and attribute disclosure in a healthcare dataset. The case study utilizes a 55500 dataset medis containing the quasi-identifiers Age, Gender, and Blood Type, as well as the sensitive attribute Medical Condition. The dataset was anonymized using k-anonymity through generalization and suppression to form equivalence classes with a minimum size of k ? 5. Subsequently, the dataset was evaluated using l-diversity to measure the diversity of sensitive attributes within each group, and t-closeness to assess the similarity between the distribution of sensitive attributes in each group and the global distribution using Earth Mover’s Distance (EMD). The results indicate that all equivalence classes satisfy k ? 5 with a suppression rate of 1.15%. The l-diversity evaluation shows that no equivalence class has l < 2, thereby minimizing the risk of attribute disclosure. The t-closeness assessment reveals that the majority of classes have EMD ? 0.15, with only one class slightly exceeding the threshold of t = 0.2. In terms of data utility, the Normalized Generalized Information Loss (NGIL) value of 0.079 (7.9%) and an AECS of 6.28 indicate a low level of information loss without over-generalization. Overall, the combination of methods successfully achieves a balance between privacy protection and data utility, ensuring that the dataset remains suitable for further analysis and secondary data publication.