Comparative Study of Variable Importance Measurements on Identification of Vulnerable Households Rini Irfani (a), Erfiani (b*), Bagus Sartono (c)
a) Department of Statistics, IPB University, Bogor, Indonesia, irfanibaehaqi[at]gmail.com
b) Department of Statistics, IPB University, Bogor, Indonesia, erfiani[at]apps.ipb.ac.id
c) Department of Statistics, IPB University, Bogor, Indonesia, bagusco[at]apps.ipb.ac.id
Abstract
Two of the seventeen goals stated in the SDGs are eradicating poverty and ending hunger. Indonesia^s poverty rate was recorded at 9.41 percent in March 2019, and the prevalence of the population with moderate or severe food insecurity was 5.42 percent. Identification of poor and food-insecure households needs to be done by using feature selection, namely information gain, gini gain, and mean decrease gini. The variables that were consistently in the top rankings were cooking fuel, the main material of the widest part of the dwelling floor, the main source of water used by the household for taking bath/washing, education status of the head of the household, the owned and the use of toilet facility, the main material of the widest part of walls, the status of household members of saver, the percentage of food expenditure, agricultural households status, and urban-rural classification. For characteristic variables that use the mean decrease gini, predictor variables with many categories tend to rank at the top for measuring variable importance. For example, variables for bathing and washing water facilities have four categories, and variables for the type of roof used to have four categories. Poor and food-insecure households have the characteristics of a house with bamboo floors, walls of woven bamboo, and cooks use charcoal and firewood.
Keywords: food insecurity, gini gain, information gain, mean decrease gini, poverty
