In the last 20 years, social media platforms have become ubiquitous. Twitter and other social media have seen exponential growth in user engagement over the years. Recent advances in ML and NLP have shown impressive results in social media analysis. This study's objective is to evaluate and compare the effectiveness of ML, specially clustering techniques, for detecting real and false news in social media based on a large-scale tweet dataset. The research focuses on optimizing clustering methods to accurately categorize tweets, enhancing the reliability of identifying misinformation on social media platforms. This study analyses 44,921 tweets to determine real and fake news detection strategies. Thorough preparation removes punctuation, spaces, and stop words to improve data quality for analysis. These models may include logistic regression, multi-layer perceptron (MLP), XGBoost, and K-means. Insightful tweet categorization is made possible by determining the ideal number of groups for K-means clustering using the Elbow approach. According to the results of the performance evaluation, K-means outperformed all other methods with regard to accuracy (99.9%), precision (100%), recall (100%), and F1-score (100%).K-means outperforms all other measures, proving that it can accurately differentiate between real and false news tweets. We will address possible dataset biases, investigate ensemble approaches and deep learning architectures to refine our models, and use sentiment analysis to improve the efficacy of false news identification in future endeavors.