This chapter investigates how different preprocessing and data balancing strategies impact fake news detection. The chapter features three experiments: (1) enriching data with auxiliary information, (2) handling informal text from social media, and (3) addressing class imbalance in Sects. 3.2, 3.3, and 3.4, respectively. Results show that enhancing input text with context improves model performance significantly. The use of preprocessing strategies like spell correction, emoji normalization, and syntactic adjustments is crucial in domains where input is noisy and brief. For data balancing, techniques like oversampling and undersampling are explored. Evaluation across various models demonstrates the importance of adapting the input pipeline to suit the task, especially in low-data environments. These insights reinforce the need for careful data preparation before applying sophisticated models.