METODE FASTICA UNTUK REDUKSI DATA DIMENSI TINGGI PADA ANALISIS SENTIMEN PARIWISATA KOTA SEMARANG MENGGUNAKAN ALGORITMA SUPPORT VECTOR MACHINE
Some communities have a voice attractions via Twitter. The opinion can be used as sentiment analysis to determine the ratings of a tourist attraction. Results of sentiment analysis is expected to assist in the improvement and evaluation of the attraction. In related research sentiment analysis previously used linear dimension reduction method, but has the disadvantage produce a linear combination of all the features that will have difficulty if dealing with data that is non-linear. Therefore, in this study used methods of non-linear dimension reduction, namely FastICA in order to improve the accuracy of Support Vector Machine classifier that can handle high-dimensional and non-linear data. This study uses the Indonesian language text contained on the social networking site Twitter. Validation is done by using a 10-Fold Cross Validation. While the measurement accuracy is measured by the Confusion Matrix and ROC curves. Results application of dimension reduction FastICA gain accuracy of 92.90% and the AUC 0.9157 which means the accuracy of 0.95% better than on Support Vector Machine itself, is proven to increase the accuracy of the SVM algorithm on the non-linier tweet data of attractions in the city of Semarang that can be classified by both in positive and negative class.