This study addressed the issue of internet disinformation by examining the identification of fake Twitter accounts. To tackle the increase in fake accounts on Twitter, detecting technologies needed to be developed. However, Traditional SVM algorithms had limitations in noisy scenarios, underperformed with more features than training data samples, and required longer training times for large datasets. To overcome these limitations and improve accuracy in recognizing fake Twitter accounts, this thesis employed a Hybrid SVM algorithm incorporating Kendall Rank Correlation, PCA, and LLE. The proposed approach recommended a Hybrid SVM algorithm, combining approaches to enhance classification performance. It used Kendall Rank Correlation to capture data correlations, PCA to reduce dimensionality, and LLE to minimize computational complexity, lowering SVM training time. After extensive testing, the proposed Hybrid SVM model demonstrated exceptional performance, achieving an accuracy and precision rate of approximately 98%. Consistent recall performance enabled reliable identification of fake accounts. The findings highlighted the effectiveness of the suggested method in spotting fake accounts and emphasized the importance of feature selection and dimensionality reduction in improving classification performance. The study contributed to social media analytics and internet security by offering insights and suggestions to address the widespread problem of fake Twitter accounts.
