Differentially Private NILM Using On-Off Matching and Naive Bayes Classification

Şimşek B., Zehir M. A., Maşazade E.

2025 Innovations in Intelligent Systems and Applications Conference (ASYU), Bursa, Türkiye, 10 - 12 Eylül 2025, ss.1-4, (Tam Metin Bildiri)

  • Yayın Türü: Bildiri / Tam Metin Bildiri
  • Doi Numarası: 10.1109/asyu67174.2025.11208501
  • Basıldığı Şehir: Bursa
  • Basıldığı Ülke: Türkiye
  • Sayfa Sayıları: ss.1-4
  • Marmara Üniversitesi Adresli: Evet

Özet

This research addresses data privacy risks in energy consumption analysis by applying Non-Intrusive Load Monitoring (NILM) techniques combined with differential privacy. Using IBM's Diffprivlib, a privacy-preserving framework was developed to detect appliance-level usage from aggregate household data. Two NILM approaches On-Off Matching with Window Shifting and Matrix Factorization were tested on real world and simulated (CREST) datasets. Gaussian Naive Bayes was used for classification. Results show that On-Off Matching offers high accuracy, especially with start/end power features, while privacy-utility trade-offs depend on the differential privacy parameter ϵ. This framework supports secure energy analytics under limited data conditions.