KAJIAN KINERJA METODE FUZZY k-NEAREST NEIGHBOR PADA PREDIKS CACAT SOFTWARE

Abstract

This research examines the model of Fuzzy k-Nearest Neighbor (Fk-NN) to predict software defects. Software defects based
on three dataset, CM1, JM1 and KC1. Datasets are derived from the promise repository. Feature selection and data
normalization applied at the stage of data pre-process. Feature selection based on Correlation-Based Feature Selection
(CFS), and data normalized based on min-max method. This research applies Fk-NN method to predict software defects.
Performance prediction consisted of five aspects: accuracy, sensitivity and precision. Testing techniques applied in this study
is 10-fold Cross Validation. To get the best performance, we applying the varied value of k and m. Range for K value is [1,
9] and m values [1.0, 1.9]. The best performance for CM1 dataset was obtained on a combination of value [k, m] = [9, 1.0].
For JM1 dataset, the best performance was obtained on a combination of value [k, m] = [9, 1.9]. For KC1 dataset, the best
performance was obtained on a combination of value [k, m] = [9, 1.5]. The results of this study indicate that the results and
performance classification with Fk-NN method highly depends on the parameters k and m, the selection of appropriate
parameters will yield the expected performance