Unwanted spam emails have become an important issue that gets in the way with user experience and email communication. These messages, which, sometimes contain harmful content and frequently promote goods or services, enter inboxes without the user’s permission. Spam detection has gotten more challenging due the rise of big data, especially in the social media space, due to the sheer amount and variety of data, which is often accompanied by noise that reduces the efficiency of analytical models. Strategies for reducing noise are effective and are essential for improving decision-making, user satisfaction, and drawing practical conclusions from vast datasets as social media platforms expand. This study utilizes the simplicity and dependability of the Naïve Bayes classifier to investigate its application in spam email detection. To enhance system performance and accuracy, the research focuses on distinguishing between spam and non-spam emails using a real-world dataset. Even in the noisy environments common in social media-based big data, the Naïve Bayes classifier demonstrates robustness in filtering spam. The classifier's performance is greatly improved by incorporating preprocessing techniques such as noise reduction, normalization, and the elimination of extraneous components like URLs, stopwords, and punctuation. In addition, the paper emphasizes how TF-IDF vectorization and the use of a simplified processing pipeline can increase the model’s operational efficiency. Furthermore, GridSearchCV's hyperparameter tuning optimizes the classifier and ensures the best outcomes. The findings indicate that the Naïve Bayes classifier can accurately identify spam in the dynamic and complex social media data environment when paired with effective preprocessing techniques. A crucial component of effective spam detection in big data, noise reduction not only highlights important patterns but also enhances model performance.