In COSMOS, we propose a novel method which combines the predictive power of Machine Learning with the processing capability of Complex Event Processing in order to predict complex events which provide the basis for Pro-active IoT6 applications. In this presentation, we highlight the technical aspects of our approach with the help of a real world use case of Intelligent Transportation System (ITS). We propose a novel prediction algorithm called Adaptive Moving Window Regression (AMWR) which uses moving window for training regression models used for prediction. The intuition of our approach is that if the input to CEP is predicted data, the complex event detected will be in future. But the prediction of dynamic IoT data streams is a challenging task which we have addressed in our work.