A New Approach to Estimating Destinations in Open Automated Fare Collection Systems based on errors-against-errors strategy

10 In transit systems, automatic fare collection systems (AFCs) are widely used. Passengers are 11 often required to use their smart cards only when entering stops, so their destination is 12 unknown. Methods have been proposed for addressing the problem, but most of those require 13 network-level AFC data. The problem remains unresolved when only one line's AFC data is 14 available. This paper tries to solve this issue for specific applications, like crowding-related 15 problems such as calculating perceived travel times. In our method, rather than minimizing 16 errors, the model is constructed so that desirable errors are produced to counter undesirable 17 errors. The task is accomplished by employing an imbalanced binary class classification 18 based on thresholding for each stop. A classification indicates whether a passenger is 19 alighting or has already alighted at the study or previous stops. Although the model may 20 produce incorrect predictions for a particular stop, it will be adjusted to make a deliberate 21 error: for every incorrect prediction of alighting, there will be a few incorrect predictions of 22 not alighting. Using this technique, we estimate how many passengers are on board the bus. 23 Our model has the functionality of an Automatic Passenger Counting (APC) system when 24 the line does not have one. 25