Internet of Things (IoT) allow data collection and sharing of many aspects of our daily life. With the huge amount of data transferred via IoT devices, privacy concerns need to be addressed while preserving the usefulness of the shared data. This research proposes a microaggregation-generative privacy preserving model for human activity recognition in IoT data. While generative deep neural networks have been used for data perturbation, data leakage and disclosing private information of the training sample is still an issue when applying the traditional deep learning generative models. We proposed a novel approach to perturb IoT data using Generative Adversarial Networks and Microaggregation while preserving the utility of its features. Our approach reduces the size of the original dataset by employing an entropy-preserving measure to discard noisy records after anonymization. The performance of the proposed approach was measured using several criteria such as classification accuracy, precision, recall, and F-score before and after anonymization. As a result, the proposed GAN-Microaggregation privacy preserving technique showed a remarkable performance in term of accuracy. Moreover, the privacy was measured, showing the benefits of the proposed approach to share IoT datasets with minimal privacy attacking surface.