Code assignment system (KASİS) for international statistical classifications

Statistical classifications have a great importance in the statistical systems of countries. Classification of economic activity, occupation, education and consumption expenditures can be given as examples for such classifications. Accuracy of studies using variables involving such classifications strongly depends on the implementation of coding correctly. As data size grows, it is usually not feasible to manually check whether the coding is error-free or not. As such, there is a clear need for an automated system to check the coding performed. In this study, a system that can be used for all statistical classifications with standard classification dictionary has been developed. The effectiveness of the proposed system has been examined using the 2013-2018 Household Budget Survey (HBS) micro data sets obtained from Turkish Statistical Institute (TURKSTAT). This data set can be regarded as the main source of the consumption expenditure statistics in Turkey. In this dataset, Classification of Individual Consumption by Purpose (COICOP) is used for classifying consumption expenditures. For each record, the codes assigned by the proposed system has been evaluated with the codes previously assigned by interviewers to observe accuracy. The proposed system distinguishes itself from supervised machine learning methods driven systems, as the former does not require training data set. The proposed system keeps learning in each iteration to exploit it in subsequent records. Also, the proposed system can be used as supporting tool for supervised machine learning methods driven systems by checking the codes assigned in the training data set. Furthermore, the proposed system can be used to assign COICOP codes for barcode data -- one of the most popular alternative data sources for Consumer Price Index (CPI).