Balancing Privacy and Accuracy in IoT Using Domain-Specific Features for Time Series Classification

$\varepsilon$
-Differential Privacy (DP) has been popularly used for anonymizing data to protect sensitive information and for machine learning (ML) tasks. However, there is a tradeoff in balancing privacy and achieving ML accuracy since
$\varepsilon\text{-DP}$
reduces the model's accuracy. Moreover, not many studies have applied DP to time series from sensors and Internet-of- Things (IoT) devices. In this work, we try to achieve the accuracy of ML models trained with
$\varepsilon\text{-DP}$
data to be as close to the ML models trained with non-anonymized data for two different physiological time series. We propose to transform time series into domain-specific 2D (image) representations such as scalograms, recurrence plots (RP), and their joint representation as inputs for training classifiers. These images allow us to apply state-of-the-art image classifiers to obtain accuracy comparable to classifiers trained on non-anonymized data by exploiting the additional information such as textured patterns from these images. To achieve classifier performance with anonymized data close to non-anonymized data, it is important to identify the value of
$\varepsilon$
and the input feature. Experimental results demonstrate that the performance of the ML models with scalograms for one of the datasets and RP for the other dataset was comparable to ML models trained on their non-anonymized versions. Motivated by the promising results, our work suggests an end-to-end IoT ML edge-cloud architecture that employs our technique to train ML models on
$\varepsilon\text{-DP}$
physiological data. Our technique ensures the privacy of individuals while processing and analyzing the data at the edge securely and efficiently.