How to embed and improve AI-ML in smart surveys for Official Statistics
Conference
65th ISI World Statistics Congress 2025
Format: IPS Abstract - WSC 2025
Keywords: sensor-data
Session: IPS 863 - Smart Survey Methodology
Thursday 9 October 10:50 a.m. - 12:30 p.m. (Europe/Amsterdam)
Abstract
Smart surveys offer new opportunities for improving social surveys, especially those based on burdensome compilation of diaries (Household Budget Survey, HBS and Time Use Survey, TUS), as they aim to exploit new data sources through personal devices (smartphones, tablets, wearables) that use sensors and provide sensor data that can supplement or substitute questionnaire data collection. How machine learning is to be used to process this smart data in order to predict survey variables for Official Statistics is the key question of this presentation. The level to which automation can be brought to replace the direct acquisition of information or replace manual processes to improve data quality and/or decrease respondent burden, is the crucial point in the use of ML in smart surveys. When results from ML models applied to sensor data can be used directly as statistical data, and when data should be fed back to respondents? How and when should training datasets be updated or improved? To improve the accuracy of the ML models, human interventions must be envisaged to assign correct labels. Also the use of external information can improve the quality of the prediction produced by ML models. Case studies discussed are in the context of HBS and TUS.