In the modern era, characterized by the widespread presence of sensor-based systems, ensuring time-series data privacy without compromising utility has emerged as a significant challenge. While current strategies for safeguarding time-series data prioritize either input data protection or privacy-preserving classification models, they often fall short in assessing the balance between data privacy and utility, particularly when strong adversary classification models are involved. Our approach introduces a novel protection technique specifically designed for multivariate Time-Series Classification. This technique involves perturbing the data by distributing the noise among features using a feature importance-based approach, thus securing the data while preserving its analytical value. We propose a dual-model evaluation system consisting of two supervised classifiers, a Privacy-Breaking Classifier and a Utility-Focused Classifier. These are designed to respectively assess the potential for privacy breaches and the extent of data utility preservation in the context of protected time-series data. The experimental results demonstrate the effectiveness and viability of our methodology. Our approach provides a framework for evaluating the privacy and utility levels of time-series data. Additionally, it guides the selection of an appropriate perturbation level to ensure both aspects are adequately addressed.