Abstract

Groups of households based on per capita expenditure is composed of two groups of poor households and non-poor households, to separate individuals or objects into a group so it can be located ata particular group can use the method of classification. The purpose of this study was to determine the classification results and errors in the results classification of households in Cilacap district based on the factors affecting the level of poverty in Cilacap with methods Multivariate Adaptive Regression Spline (MARS). MARS is a nonparametric regression method that can be used for high-dimensional data is. To get the best MARS models, do a combination of value Basis Function (BF), Maximum Interaction (MI), and the Minimum Observation (MO) by trial and error. The best model is the model that is used in combination with BF = 45, MI = 3, MO = 1 because it has the smallest value that is equal to 0,030 GCV. Based on the variables that affect groups of households in Cilacap, the result of classification of 37 households with poor category, 34 households appropriately classified as poor, while one 3 households are classified as poor. Likewise, of the 113 households with non-poor category, 113 households appropriately classified into the category of not poor, and no household misclassified into the household with non-poor category. Retrieved classification accuracy of 98.00% of the value of Apparent Error Rate (APER) at 2.00% and the Press's Q test showed that statistically MARS method has been consistent in classifying the data. So as to further research on the classification suggested using the method MARS, by looking at the value of the smallest GCV and GCV value if they have the same smallest it can be seen with the smallest MSE value judgment.