Wireless Sensor Networks (WSNs) are fundamental blocks of Internet of Things (IoT) which made it to proliferate into many real-time diversified applications. Typically, WSN nodes are small and battery powered devices. Hence, energy efficient data aggregation method which maximizes the network lifetime is the paramount importance. Compressed Sensing (CS) is one of the approaches proves to be very promising for energy efficient data aggregation in WSN. The sparsity of the sensor data can change significantly due to their time varying nature and it affects the recovery of the measured signal. Most of the existing CS aided data aggregation methods for WSN are neither optimized nor address the issue of change in data sparsity, which demands additional energy to maintain low reconstruction accuracy. Thus, variation in the sparsity of the sensor data makes data aggregation to be energy inefficient that inherently reduces network lifetime. To overcome this, we propose an optimal method for CS-WSN where we use a dictionary as a sparsifier that learns with adaptive sparse training data. The proposed method achieves minimum reconstruction error by accounting changes in the data sparsity while data can be delivered to the sink with low energy consumption thereby improving energy efficiency. © 2018 IEEE.