The infiltration wells program in Jakarta,Indonesia, is one of the issues that has become a hot topic on Twitter after a political figure's car fell into one of the infiltration wells in South Jakarta. As a result, a growing number of people have spoken out about the program's benefits and drawbacks, which later cause pros and cons. This study aims to determine public sentiment on Twitter about the Jakarta infiltration wells program and to determine the accuracy and performance of the Naive Bayes, Support Vector Machine, and K-Nearest Neighbor as the classification algorithms used in this research. With SMOTE, balanced data of 591 positive and 591 negative tweets was obtained, with testing data of 138 tweets. The result shows the highest accuracy of 93.32 percent, as well as high performance was reached with SVM, followed by Naive Bayes in second place, and KNN in third place. The result of this study also finds that most of the tweets have negative sentiments, mostly about the program inability to handle floods, the formation of puddles and damages on roads, high allocation program budgets, and protests of residents who had not been compensated for their assistance in building the infiltration wells.