Predicting Bus Arrivals Using One Bus Away Real-Time Data

6 Providing real-time bus arrival times can enhance the experience of riders using 7 public transit, but errors in the predictions can cause a decrease in ridership. 8 Using data collected from a local transit agency, we created models to predict 9 arrival times using K-Nearest Neighbors and Kernel Regression and seven sets 10 of features. The best feature set had a RMSE of 34 seconds, a large 11 improvement over the baseline of always predicting on-time arrivals, which had 12 an RMSE of 99 seconds. 13 14 15