The multi-sensor-based human behavior recognition system is proposed to improve recognition rate of system on human behavior

and can reduce system energy consumption. Data features are extracted from sensor data information by mathematical statistics means within sliding time window; by using weka software

analyze data festures

a two-layer classification recognition algorithm based on decision tree is designed to recognize eight common human behaviors. Experimental result demonstrates that this system achieves high recognition rate of93. 12 %

and reduces system energy consumption

and it is portable and has strong practical applicability.