The autonomy of mobile devices presents a major research challenge. The energy consumption is due to hardware and software components. The optimization of this consumption cannot be ensured without identifying the sources of this consumption. The first aim of this work is to select the attributes that strongly influence energy efficiency from a set of attributes based on a large raw data set. The second objective is to propose new formulas and rules of associations which connect the energy by the selected attributes to track the change in power consumption in a mobile device. To carry out this work, we used the techniques of machine learning and data mining based on real experimental data collected in the scientific experiment of the 'Tour de France'. This database contains 52, 056 rows of information and 20 attributes, allowed us to conduct an effective study on our raw data set. The results found by our approach are encouraging. Indeed, we have developed relevant rules for the energy variation in a mobile device.