SEMIVARIOGRAM ISOTROPIC ROBUST AND NONROBUST ANALYSIS IN BANDUNG’S THEFT DATA

• Main Author: Trismayangsari, Riska
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/44303
• Institution: Institut Teknologi Bandung

In this study loss value caused by theft in Bandung will modeled by semivariogram isotropic robust and non robust. Semivariogram can represent spatial correlation between the theft locations by variance of difference loss of pair locations. The data have two big loss value that called outliers. Therefore, in this study used two approach robust semivariogram that are Cressie-Hawkins and Dowd also non robust that is Matheron. Determination of the best model in semivariogram can be done by determine the right parameters model nugget effect (????0), sill (????), and range (????). In this study ???? will determined based on variance of loss value and statistic of experimental semivariogram (?????(?)) that are mean (??????(?)), first quartile (????1(?????(?))), median(????2(?????(?))), and third quartile(????3(?????(?))). Through determination the ???? will estimated ???? by numeric method. The result shows that from three semivariogram approachs the right model for this data is Gauss with ???? around 2 kilometers. It said that the case of theft which happened in radius 2 kilometers has similar loss value. Data that has big loss or outliers has candidate of sill is median of experimental semivariogram. However, data that loss value is similar or homogeneous has candidate of sill are variance of loss and third quartile of experimental semivariogram. Furthermore, in this studies obtained value of ???? = 2. It indicates that the case of theft that happened in radius 2 kilometers has similar loss value. Furthermore, the result also said that ???? of outlier is (????2(?????(?))). Nevertheless, data that has low variability of loss can choose ???? are variance of loss and (????3(?????(?))). Furthermore, after has the best model the next step is kriging as validation. The kriging is adapt the process of Jackknife that is removed one location then estimate the loss value at that lovation. Repeat the step until all locations has estimate value. The last is calculate Sum Square Error (SSE) of the real and estimated loss value.