Developing a Robust Median and MAD-Based Estimator for Casewise and Cellwise Outlier Detection in South Sulawesi Socio-Economic Data

Abstract

Socio-economic data analysis at the regional level frequently faces problems of spatial heterogeneity and extreme variability that give rise to outliers. Traditional approaches using casewise deletion often discard valuable information entirely just because of anomalies in a small fraction of attributes. This study proposes the development and application of a robust estimator based on Median and Median Absolute Deviation (MAD) to detect outliers using a cellwise paradigm (data cell-based) and compares it with a casewise approach (observation row-based) on socio-economic datasets of Regencies/Cities in South Sulawesi Province. The analyzed indicators include the Human Development Index (HDI), Regional Gross Domestic Product (RGDP), Percentage of Poor Population, Per Capita Expenditure, and the Open Unemployment Rate (OUR). Experimental results show that the robust estimator with a threshold of 2.24 is able to map cellwise outliers accurately without reducing the dimension of the observation data. Heatmap and Robust Z-Score Scatter Plot visualizations reveal that Makassar City is an extreme outlier in the RGDP indicator, while other anomalies are found in the OUR indicator in the Enrekang, Bulukumba, and Bone regions. Ultimately, this approach proves to be superior in maintaining the integrity of macroeconomic datasets compared to classical methods.