A new approach to Data Fusion through Constrained Principal Component Analysis

Data Fusion consists of merging information coming from two different surveys. The first is called “reference” or “donor survey” while the second is called “punctual” or “receptor survey”. The aim is to complete the receptor matrix exploiting information acquired from the donor matrix. The two independent surveys have a block of common variables used as a bridge between them. In this work a Data Fusion methodology based on the Constrained Principal Component Analysis (CPCA) technique is presented. The proposed method allows to impute the missing information into the second survey taking into account knowledge about non-symmetric relationship structure among variables.