Internet of Things (IoT) sensors continuously generate large volumes of data that must be transmitted to data centers. The resulting high bandwidth demand calls for methods to reduce the number of data points while preserving data quality. In this context, the DREAM (Dynamic Regression Algorithm) has shown that the original data points can be represented by a smaller number of samples while maintaining acceptable accuracy. This paper presents an application of DREAM in an intelligent IoT data collection system. IoT devices fit linear regression models over recent measurements and continuously evaluate the coefficient of determination to decide when data should be transmitted. At the beginning of each segment, two data points are sent to establish the trend of the data. When the coefficient of determination remains above a predefined threshold, no additional data are transmitted; when the coefficient of determination falls below the threshold, the previous data point is sent to mark the end of the current segment. The missing values are then reconstructed at the backend by linear interpolation between consecutive transmitted points. A system of two nodes (indoor and outdoor) collecting temperature and humidity data using different sampling periods is used to compare the proposed approach and the baseline method in sending all data in terms of data sending reduction. The results show that DREAM can reduce the number of transmitted data points while maintaining acceptable accuracy.