Social media platforms, particularly Twitter, have become essential sources of data for various applications, including marketing and customer service. This study focuses on analyzing customer interactions with Amazon's official support account, "@AmazonHelp," to understand and predict changes in customer sentiment during these interactions. Using the Twitter API, we extracted English-language tweets mentioning "@AmazonHelp," pre-processed the data, and categorized conversations to facilitate analysis. The primary objectives were to classify changes in customer sentiment and predict the overall sentiment change based on initial sentiment. We conducted experiments using multiple machines learning algorithms, including K-nearest neighbor, Naive Bayes, Artificial Neural Network, Bayes Net, Support Vector Machine, Logistic Regression, and Bagging with RepTree. Our dataset comprised over 6,500 conversations, filtered to include those with four or more tweets. Results indicated that K-nearest neighbor and Support Vector Machine offered the best balance between accuracy and F-measure, while Bagging with RepTree achieved the highest accuracy but had a lower F-measure. This study demonstrates the potential of integrating sentiment analysis and machine learning to effectively predict customer sentiment in social networks, providing valuable insights for improving customer engagement strategies.