A Privacy Preserving Anomaly Detection Framework for Cooperative Smart Farming Ecosystem

The agriculture sector has seen growing applications of AI and data intensive systems. Typically, individual farm owners join together to form agricultural cooperatives to share resources, data, and domain knowledge. These data intensive cooperatives help generate AI supported insights for member farmers. However, this leads to a rising concern among individual smart farm owners about the privacy of their data, especially while sharing the data with the co-op. In this paper, we present a framework where the individual smart farm owner's privacy is preserved, as it is shared to train robust anomaly detection models at the cooperative level. Here, we preserve the privacy of each farm owner by adding noise to their data through data perturbation techniques such as white Gaussian noise. Our experimental results show that the anomaly detection models can identify various anomalous events even when the training data is transformed with white noise. Further, we evaluate our framework and compare the detection performance on non-transformed and transformed data that belongs to multiple smart farms present in a cooperative.