Enriquez, Janina Marella G.; Simbahan, Kyle Patrick G.
Hybrid support vector machine algorithm for twitter fake account detection
- Undergraduate Thesis: (Bachelor of Science in Computer Science) Pamantasan ng Lungsod ng Maynila, 2023.
ABSTRACT: This study addresses the growing issue of internet disinformation by providing an in-depth examination of identifying face Twitter accounts. To effectively address the increase in the number of fake accounts on Twitter, capable detecting technologies must be developed. However, Traditional SVM algorithms have limits in scenarios with high levels of noise, underperform when there are more features than that of the number of training data samples, and require longer training times when dealing with large datasets. To solve these limitations and obtain increased accuracy in recognizing fake Twitter accounts, this thesis employed a Hybrid SVM algorithm that incorporates Kendall Rank Correlation, PCA, and LLE. The proposed approach recommended a Hybrid Support Vector Machine (SVM) algorithm, which combined numerous approaches to improve classification performance. The approach used Kendall Rank Correlation to capture correlations between the data, Principal components analysis (PCA) to reduce dimensionality, and local linear embedding (LLE) to further minimize the computational complexity of the dataset, lowering the training time of the SVM classifier. The proposed Hybrid model exhibits extraordinary performance after lengthy testing and evaluation, obtaining an exceptional accuracy and precision rate of around 98%. Since the recall’s performance is consistent, face accounts may be reliably identified. The study’s findings demonstrate not only the extent to which the suggested method works for spotting fake accounts, but also how crucial feature selection and dimensionality reduction are for improving classification performance. The study makes a contribution to the fields of social media analytics and internet security by offering insightful information and useful suggestions for dealing with the widespread problem of fake Twitter accounts.
5
ABSTRACT: This study addresses the growing issue of internet disinformation by providing an in-depth examination of identifying face Twitter accounts. To effectively address the increase in the number of fake accounts on Twitter, capable detecting technologies must be developed. However, Traditional SVM algorithms have limits in scenarios with high levels of noise, underperform when there are more features than that of the number of training data samples, and require longer training times when dealing with large datasets. To solve these limitations and obtain increased accuracy in recognizing fake Twitter accounts, this thesis employed a Hybrid SVM algorithm that incorporates Kendall Rank Correlation, PCA, and LLE. The proposed approach recommended a Hybrid Support Vector Machine (SVM) algorithm, which combined numerous approaches to improve classification performance. The approach used Kendall Rank Correlation to capture correlations between the data, Principal components analysis (PCA) to reduce dimensionality, and local linear embedding (LLE) to further minimize the computational complexity of the dataset, lowering the training time of the SVM classifier. The proposed Hybrid model exhibits extraordinary performance after lengthy testing and evaluation, obtaining an exceptional accuracy and precision rate of around 98%. Since the recall's performance is consistent, face accounts may be reliably identified. The study's findings demonstrate not only the extent to which the suggested method works for spotting fake accounts, but also how crucial feature selection and dimensionality reduction are for improving classification performance. The study makes a contribution to the fields of social media analytics and internet security by offering insightful information and useful suggestions for dealing with the widespread problem of fake Twitter accounts.