Improving classification performance of public complaints with TF-IGM weighting: Case study : Media center E-wadul surabaya

摘要

Currently Media Center e-Wadul still uses manual labeling in the process of complaint submission. As a result, Media Center administration takes a long time in coordinating with regional work unit (SKPD) to respond to complaints registered. Therefore, it is necessary to classify complaints based on SKPD to speed up the timing of complaint submission. The challenge of classification using text data is to have a high dimension due to a large number of features. In addition, features that appear in almost all classes and even all classes and do not characterize a class are challenges in this research. The proposed term weighting is Term Frequency-Inverse Gravity Moment (TF-IGM). TF-IGM can calculate distinguishing class precisely of a term especially for multiclass problems in this study. The famous Term Frequency-Inverse-Document Frequency (TF-IDF) and TF-Binary weighting methods are also used as a comparison. The classification is performed on Support Vector Machine (SVM), Naive Bayes and K-Nearest Neighbor (KNN) algorithm. In this research, the incoming public complaints will be processed through the pre-process stage, term weighting stage, and classification stage. The classification performance using TF-IGM weighting on SVM method yielded the best value compared to others with accuracy, precision, recall and f-measure respectively 80.11%, 80.70%, 80.10%, and 80.20%.