Because of the huge number of private cars going through the streets of Jakarta, traffic congestion develops regularly, prompting the Provincial Government to establish TransJakarta. Often the TransJakarta users wish to ask questions, file complaints or add suggestions to TransJakarta via Twitter. To make it easier and faster for TransJakarta to respond to tweets, it is vital for them to understand the categories of tweets. In order to do this, Tweet categories were determined using data collected from the Twitter API. The text preprocessing was done first then proceeded with calculating and weighting each word using Term Frequency-Inverse Document Frequency (TF-IDF). In addition, Genetic Algorithm (GA) was proposed to be used in feature selection. K-means and Density-Based Spatial Clustering of Applications with Noise (DBSCAN) methods are compared based on the silhouette coefficient value to determine the categories of tweets and then visualized using word clouds. The clustering results show that the best method is DBSCAN with GA-based feature selection because it produces a high silhouette coefficient value with less noise than without GA-based feature selection. Clustering obtained four categories of tweets, namely bus stop/route, bus facilities, bus cleanliness, and TransJakarta's consistency.