A New Method To Match Data Set Applied To Electric Market

The French household electricity market currently dominated by EDF, is going to be opened up to free market economy. In this context, a better understanding of its customers behaviour would be a key feature for the success of the company. To achieve this goal, EDF holds two information sources i.e., a comprehensive customers' invoicing file with few individual data though, as well as survey results performed in regional centres, which contain more data per customer. We herein present a new method of data fusion based on the generation of virtual individuals. For each of them, a full set of variables will be accessible. This procedure is based on two steps. Firstly, a Multiple Correspondences Analysis (MCA) from an existing master sample is performed on fundamental variables. A sample of virtual individuals (SVI) is randomly selected, based on the distribution of each significant MCA factor. Then, for each virtual individual, a specific value is given for each fundamental variable which is the most correlated to one of the MCA factors. Secondly, the distinct sets of secondary samples are grafted to the previous SVI. A simultaneous estimation of variables distribution is made as based on PLS regression on variables shared by all samples. The use of this method brings about two advantages, namely the possibility to choose SVI size and the avoidance of variance underestimation generally encountered in using the imputation methods. This process has been so far applied to the treatment of two databases, i.e. two surveys, in order to generate the expected artificial sample. Validation and perspectives will be herein further discussed.