TY  - THES
N1  - Pembimbing: Dr. Agung Fatwanto, S.Si., M.Kom.
ID  - digilib67437
UR  - https://digilib.uin-suka.ac.id/id/eprint/67437/
A1  - Agus Suparman, NIM.: 20206051008
Y1  - 2024/08/19/
N2  - Social media has evolved as a medium for the mass dissemination of information (including fake news). Fake news on social media platforms like Twitter/X often gets retweeted quickly and spreads widely, sometimes even going viral. Therefore, a mechanism is needed to automatically classify fake news on social media. This study aims to compare the performance of two machine learning models, namely Feed Forward Neural Network (FFNN) and Support Vector Machine (SVM), in automatically classifying fake news. Additionally, this research intends to compare the impact of the test data proportion relative to the training data, as well as the effect of class balance in the dataset on classification performance. This study employs an experimental quantitative approach. The experiment process involved training and testing the FFNN and SVM models using a dataset of Twitter/X post metadata, including the number of followers, number of quote tweets, number of retweets, and number of likes. The data used for the experiment totaled 2,104 samples collected through Astramaya by DroneEmprit, with an additional 1,781 samples obtained from the TurnBackHoax site. The classification results of both models were evaluated using a Confusion Matrix to calculate accuracy, precision, recall, and F1-score. The findings show that the SVM model generally achieved higher accuracy, recall, and F1-score compared to the FFNN model, which only excelled in precision. Both models generally experienced a decline in performance with an increasing proportion of test data (except for a 40% test data proportion, which performed better than a 20% test data proportion). Both models demonstrated better classification performance when trained on a dataset with balanced class
proportions, as opposed to an imbalanced dataset or one balanced using the SMOTE Oversampling technique. The SVM model produced better classification results than the FFNN model. The performance of both models was also influenced by the proportion of training and test data. In general, the higher the proportion of training data, the better the performance of both models. Moreover, the balance of class proportions in the dataset used for training and testing also impacted the models' classification performance. Overall, both models performed better when trained on a dataset with naturally balanced class proportions.
PB  - UIN SUNAN KALIJAGA YOGYAKARTA
KW  - Klasifikasi Berita Palsu
KW  -  Feed Forward Neural Network
KW  -  Support Vector Machine
KW  -  Twitter/X
KW  -  Machine Learning
M1  - masters
TI  - ANALISIS KINERJA FEED FORWARD NEURAL NETWORK DAN SUPPORT VECTOR MACHINE DALAM MENGKLASIFIKASIKAN BERITA PALSU DI TWITTER/X
AV  - restricted
EP  - 93
ER  -