OUTLIER DETECTION MODEL IN SOCIAL ECONOMIC DATA, CASE STUDY : NATIONAL SOCIAL ECONOMIC SURVEY.

• Main Author: Ivander, Muniaga
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/85213
• Institution: Institut Teknologi Bandung

Social Economic Survey used by government and researches to perform many calculations such as consumer price index, and poverty measurement. The quality of this survey data is so important that small amount of falsifying act conduct by enumerator in the process of gathering the data can have serious impact. The act of falsifying the data can be such as enumerator falsify part or all of the interview content, deliberately miscoding a question to avoid follow-up questions and enumerators did not go in depth with the questions so that respondents did not provide relevant answer. Outlier detection used by many researches to detect this falsifying act. Outlier approach by this study not as noise that should be removed rather but as observation that slightly different from normal behavior of data that result from falsifying the data. Local Outlier Factor use in this study to perform labelling the data between outlier and inlier, we use parameter of LOF such as MinPTS (LB)=10 and threshold of outlier data greater than 2. After the data has been labelled, 3 supervised algorithm use to perform predictive outlier use such as Naïve Bayes, Random Forest and SVM. The results shows that SVM algorithm give better value in accuracy(98.67 ) and precision (99,49).