Step 1

Select the desired transaction from the database according to the criteria (by conductor, driver, trip, ect.) and sort by date and time.

Step 2

Make a series of time intervals between transactions. Cluster the resulting series and take the desired data cluster.

Step 3

Calculate the values of interest (for example, the average time between neighboring transactions, the minimum time between transactions, the median time between transactions)

Step 4

Draw a histogram of the resulting series and analyze the obtained properties.

To calculate the performance of conductors, we use the __JoinPAY Transport__ ERP cloud system, which allows you to collect a large amount of different information needed to build an evaluation model.

As criteria for the efficiency of the conductor, we use not only the usual integral characteristics (such as, for example, the average amount of money brought per trip), but also differential characteristics based not on the number of transactions per trip as a whole, but on the time intervals between neighboring transactions.

They are determined only by the speed of movement of the conductor inside the cabin and the speed of their work with the terminal, i.e. how quickly and clearly the conductor tears off the check, how rarely the contactless payments fail, etc.

Thus, the main series of data from which the information will be extracted is a series of time intervals between consecutive transactions of the conductor, which is easily obtained from the array of transaction times stored in the system.

Find out more information about calculating efficiency in our blog.

Here is an example of the simulation of the conductor’s work, carried out in the interests of the public transport company to identify the key statistical characteristics of the transactions time series in order to determine the degree of fraud of the employee.

It is well known that a good model should preserve the most important features of the studied phenomenon, which correctly represent the desired properties of the studied process. And at the same time it is necessary to exclude insignificant features from the model for its maximum simplification. And the question of what to leave and what to exclude is one of the most important in the preparation of such models. It is solved by brute force search - let's try to simplify the model as much as possible and compare the simulation results with reality. If they reproduce the main aspects of this reality, then it is possible, firstly, to try to determine the properties of the fraud and, secondly, to make the model more complicated for a more accurate reproduction of the real time transactions series.

Fraud is modeled as follows – in the case of a random cash transaction, it is not fixed in the transaction time array. This is equivalent to the fact that the ticket from the previous cash transaction, which the passenger did not receive on hand for some reason, is immediately transferred to the passenger with the next cash payment.

Note that even such a relatively simple model allowed identifying metrics that clearly indicate the presence of fraud in the transaction series of a particular employee.