Aggregating data using count-based windows is a formidable challenge in large-scale, decentralized IoT networks, where events arrive at varying rates, unordered, and at diverse times. Existing methods are either centralized and slow, or decentralized with potential inaccuracies. This paper introduces Deco, an approach for fast and accurate aggregation of count-based windows in decentralized stream processing that uses a lightweight prediction method to derive local window sizes from previously observed event rates, applying corrections when necessary.