An Artificial Neural Network Model for Evaluating Workers' Performance at Tollbooths

Abstract

The Brazilian roadway concessions program significantly increased the number of toll plazas in highways. More than 65% of Brazilian tollbooths utilize manual toll collection. Manual toll collection causes most problems related to toll charging, i.e.: delays experienced by drivers, and costs of operating staff. The evaluation of the performance of tolling staff in manual toll collection is an important factor to increase service levels in these infrastructures. This paper presents an Artificial Neural Network forecasting model for service times at manual tollbooths. The technical literature shows that many factors influence service times. Data was collected at 13 toll plazas in the state of Rio Grande do Sul (Brazil), representing over 36% of the implemented toll plazas in the state. Artificial Neural Networks were used to construct three models: a model to predict the minimum service times, a model for maximum service times and a third model to predict the 85th Percentile. The models aim to analyze the variability of service times for the same group of input variables. The sensibility analysis of the models indicates that service times are highly dependent on traffic flows at toll plazas. Average service time is usually used to evaluate toll workers' performance but average service time cannot capture the variability of important factors involved in tolling operations. This model was used to evaluate the performance of manual toll workers, providing a sensitive procedure for performance evaluation.