The proliferation of online reviews has necessitated the development of effective methods for detecting fake reviews, which can mislead consumers and undermine trust in digital platforms. This research explores an advanced approach to fake review detection by leveraging a combination of sentiment analysis, readability metrics, and topic modelling features. We utilize a comprehensive feature set, including the RoBERTa score, VADER pos, average word length, textual content, and sentiment scores derived from emojis. In the preprocessing stage, emojis are extracted and assigned sentiment scores based on a sentiment dictionary, enhancing the emotional context of the reviews. The integrated dataset is subsequently classified using a Support Vector Machine (SVM) classifier, which is particularly effective in handling high-dimensional data. The SVM model identifies the optimal hyperplane to separate real and fake reviews, utilizing the fused feature set to capture subtle patterns and relationships within the data. The performance of the classifier is evaluated using accuracy, demonstrating significant improvements in detection capabilities. This study highlights the importance of a holistic approach to fake review detection, combining various features to enhance the reliability of sentiment analysis. The findings indicate that integrating sentiment scores from emojis with traditional text features can substantially improve classification accuracy, ultimately contributing to more robust and trustworthy online review systems.