<![CDATA[Across domains like healthcare, finance and environmental monitoring time series forecasting has many applications and often deals with sensitive data. Although ARIMA models are popular due to high interpretability and accuracy and the basic form of ARIMA does not need a large amount of data for modeling, it requires plaintext data access which raises problems in cloud computing with privacy. Homomorphically Evaluating the Autoregressive Component of ARIMA for Privacy-Preserving Forecasting through CKKS-based Fully Homomorphic Encryption (FHE) in the TenSEAL library A temperature dataset was used to train an ARIMA(2,1,1) model with the autoregressive coefficients in plaintext and more specifically, as part of plaintext and encrypted workflows. The CKKS Model Recent observations were encrypted with CKKS thus allowing forecasting on ciphertext without ever exposing the data. The outcomes indicate, that the prediction using ciphertext matched pretty well to the one with plaintext and had an absolute error of 0.000001 along with accuracy of 99.995%. Our findings validate that not only does CKKS-based FHE retain forecasting accuracy in a secure manner and enable the combination of FHE with classical forecasting models, but also gives rise to future work on encrypted ARIMA modelling, multivariate forecasting as well as privacy-preserving machine learning.]]>
Copyrights © 2026
