Analisis Perbandingan Data Sintesis Privacy-Preserving Menggunakan CTGAN dan Metode Differential Privacy untuk Pelatihan Model LSTM dalam Klasifikasi Deteksi Jatuh pada Lansia
Abstrak
Jatuh pada lansia menjadi salah satu penyebab masalah cedera fatal dan kematian di dunia. Pertumbuhan populasi lansia yang signifikan menyebabkan tingginya angka prevalensi jatuh dan mendorong kebutuhan solusi deteksi dini jatuh pada lansia. Pengembangan Fall Detection Systems (FDS) menjadi jawaban karena FDS dapat menawarkan respons yang cepat ketika kejadian jatuh terjadi. Namun, pengembangan FDS memiliki tantangan berupa keterbatasan data, ketidakseimbangan kelas, dan risiko pelanggaran privasi. Penelitian ini bertujuan untuk membandingkan dua metode privacy-preserving, yaitu model CTGAN dan DP. Penelitian diawali dengan implementasi FDS berbasis Internet of Things (IoT) dan mengumpulkan data aktivitas lansia. Lalu, melakukan pengukuran utilitas statistik dengan Kolmogorov–Smirnov (KS) complement dan Wasserstein distance, pengukuran aspek privasi dengan Nearest Neighbor DIstance Ratio (NNDR), dan pengukuran utilitas model dengan skenario Train-on-Synthetic, Test-on-Real (TSTR). Hasil menunjukkan bahwa CTGAN menjadi metode privacy-preserving yang lebih baik daripada DP, pada utilitas statistik, CTGAN mengungguli DP dengan nilai KS complement di 0.902 dan Wasserstein distance di 0.329, dibandingkan dengan DP di 0.861 dan 0.690. Pada segi privasi, CTGAN mengungguli DP di persentil ke-5 dengan nilai 0.63 dibandingkan dengan 0.57 pada DP. Terakhir, dari segi utilitas model, CTGAN mengungguli DP dengan nilai akurasi dan F1-score yang lebih tinggi dari DP. Kata kunci: conditional tabular generative adversarial network, differential privacy, fall detection, long short- term memory, privacy-preserving
Unduhan
Referensi
[1] R. Patel dan J. E. Gallagher, “Healthy ageing and oral health: priority, policy and public health,” 1 Desember 2024, Springer Nature. doi: 10.1038/s41405-024-00262-z.
[2] Y. Chen dkk., “Global, regional, and national burden of falls among older adults: findings from the Global Burden of Disease Study 2021 and Projections to 2040,” npj Aging, vol. 11, no. 1, Des 2025, doi: 10.1038/s41514-025-00275-4.
[3] Z. Mohammad, A. R. Anwary, M. F. Mridha, M. S. H. Shovon, dan M. Vassallo, “An Enhanced Ensemble Deep Neural Network Approach for Elderly Fall Detection System Based on Wearable Sensors,” Sensors, vol. 23, no. 10, Mei 2023, doi: 10.3390/s23104774.
[4] A. Butt, S. Narejo, M. R. Anjum, M. U. Yonus, M. Memon, dan A. A. Samejo, “Fall Detection Using LSTM and Transfer Learning,” Wirel. Pers. Commun., vol. 126, no. 2, hlm. 1733–1750, Sep 2022, doi: 10.1007/s11277-022-09819-3.
[5] N. Verma, S. Mundody, dan R. M. R. Guddeti, “An Efficient AI and IoT Enabled System for Human Activity Monitoring and Fall Detection,” dalam 2024 15th International Conference on Computing Communication and Networking Technologies, ICCCNT 2024, Institute of Electrical and Electronics Engineers Inc., 2024. doi: 10.1109/ICCCNT61001.2024.10725908.
[6] A. I. K. Kalupahana, A. N. Balaji, X. Xiao, dan L. S. Peh, “SeRaNDiP - Leveraging Inherent Sensor Random Noise for Differential Privacy Preservation in Wearable Community Sensing Applications,” Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., vol. 7, no. 2, Jun 2023, doi: 10.1145/3596252.
[7] A. El Ouadrhiri dan A. Abdelhadi, “Differential Privacy for Deep and Federated Learning: A Survey,” IEEE Access, vol. 10, hlm. 22359–22380, 2022, doi: 10.1109/ACCESS.2022.3151670.
[8] A. Kiran dan S. S. Kumar, “A Methodology and an Empirical Analysis to Determine the Most Suitable Synthetic Data Generator,” IEEE Access, vol. 12, hlm. 12209–12228, 2024, doi: 10.1109/ACCESS.2024.3354277.
[9] E. Callahan dan L. MacNeill, “Synthetic Tabular Data Generation for Privacy-Preserving Machine Learning. Journal of Computer Science and Software Applications,” Journal of Computer Science and Software Applications, vol. 5, no. 7, 2025.
[10] J. Zhang, J. Li, dan W. Wang, “A class-imbalanced deep learning fall detection algorithm using wearable sensors,” Sensors, vol. 21, no. 19, Okt 2021, doi: 10.3390/s21196511.
[11] S. Alabdulwahab, Y. T. Kim, A. Seo, dan Y. Son, “Generating Synthetic Dataset for ML-Based IDS Using CTGAN and Feature Selection to Protect Smart IoT Environments,” Applied Sciences (Switzerland), vol. 13, no. 19, Okt 2023, doi: 10.3390/app131910951.
[12] O. Parise, R. Kronenberger, G. Parise, C. de Asmundis, S. Gelsomino, dan M. La Meir, “CTGAN-driven synthetic data generation: A multidisciplinary, expert-guided approach (TIMA),” Comput. Methods Programs Biomed., vol. 259, Feb 2025, doi: 10.1016/j.cmpb.2024.108523.
[13] H. Park, H. Mitsumine, dan M. Fujii, “Automatic Adjustment of the Distance Ratio Threshold in Nearest Neighbor Distance Ratio Matching for Robust Camera Tracking,” IEICE Transactions, vol. 94-D, hlm. 938–940, Apr 2011, doi: 10.1587/transinf.E94.D.938.
[14] J.-W. Lee, J.-H. Jeong, J.-W. Kang, S.-I. Na, dan D.-S. Jeong, “Image Feature Point Selection Method Using Nearest Neighbor Distance Ratio Matching,” Journal of the Institute of Electronics and Information Engineers, Apr 2012, doi: 10.5573/ieek.2012.49.12.124.
[15] A. P S, S. Kumar, dan V. K R, “MQTT Implementations, Open Issues, and Challenges: A Detailed Comparison and Survey,” International Journal of Sensors, Wireless Communications and Control, vol. 13, Apr 2022, doi: 10.2174/2210327913666221216152446.
[16] Y. J. Park, S. Y. Jung, T. Y. Son, dan S. J. Kang, “Self-Organizing IoT Device-Based Smart Diagnosing Assistance System for Activities of Daily Living,” Sensors, vol. 21, no. 3, 2021, doi: 10.3390/s21030785.
[17] S. Watanuki dkk., “Applying a Method for Augmenting Data Mixed from Two Different Sources Using Deep Generative Neural Networks to Management Science,” Applied Sciences, vol. 14, no. 1, 2024, doi: 10.3390/app14010378.
[18] Z. Zhao, A. Kunar, R. Birke, H. der Scheer, dan L. Y. Chen, “CTAB-GAN+: enhancing tabular data synthesis,” Front. Big Data, vol. Volume 6-2023, 2024, doi: 10.3389/fdata.2023.1296508.
[19] M. Khavkin dan E. Toch, “Investigating the impact of differential privacy obfuscation on users’ data disclosure decisions,” Decis. Support Syst., vol. 196, hlm. 114474, 2025, doi: https://doi.org/10.1016/j.dss.2025.114474.
[20] M. Pereira, M. Kshirsagar, S. Mukherjee, R. Dodhia, J. L. Ferres, dan R. de Sousa, “Assessment of differentially private synthetic data for utility and fairness in end-to-end machine learning pipelines for tabular data,” PLoS One, vol. 19, no. 2 February, Feb 2024, doi: 10.1371/journal.pone.0297271.
[21] J. Fabuyi, “Leveraging Synthetic Data as a Tool to Combat Bias in Artificial Intelligence (AI) Model Training,” Journal of Engineering Research and Reports, vol. 26, hlm. 24–46, Apr 2024, doi: 10.9734/jerr/2024/v26i121337.
[22] Y. He, R. Vershynin, dan Y. Zhu, “Algorithmically Effective Differentially Private Synthetic Data,” dalam Proceedings of Thirty Sixth Conference on Learning Theory, G. Neu dan L. Rosasco, Ed., dalam Proceedings of Machine Learning Research, vol. 195. PMLR, Apr 2023, hlm. 3941–3968. [Daring]. Tersedia pada: https://proceedings.mlr.press/v195/he23a.htm l
[23] Q. Liu, M. Khalil, J. Jovanovic, dan R. Shakya, “Scaling While Privacy Preserving: A Comprehensive Synthetic Tabular Data Generation and Evaluation in Learning Analytics,” dalam ACM International Conference Proceeding Series, Association for Computing Machinery, Mar 2024, hlm. 620–631. doi: 10.1145/3636555.3636921.
