In the domain of crisis management for telecommunications infrastructures, the autonomous detection of cell outages within cellular networks is of paramount importance for prompt identification and resolution in ensuring uninterrupted connectivity to users. Traditional methods usually involve data aggregation at the core network, which is responsible for identifying cell failures. Proposing a novel approach, we leverage a Machine Learning-based distributed and cooperative feature removal mechanism in order to preserve the privacy of data and avoid any degradation in classification performance. Simulations carried out on a dataset retrieved from a real 4G-LTE rollout demonstrate that the proposed approach, through a cooperation among agents, maintains or even slightly improves accuracy, precision, recall, and F1-score in outage prediction compared to other conventional methods, showcasing its efficacy for cell outage detection purposes while maintaining data privacy.