A weighted k-NN based approach for corridor level travel-time prediction

The main focus of this research is predicting travel-time for an arterial corridor using the improvised k nearest neighbor model (k-NN). This study aims at improving the prediction accuracy of the k-NN model by optimizing the k value used in the modelling stage. The study proposes a k-Tree based k-NN method to learn an optimal k value for different test samples. In the training stage, the k-Tree method first learns an optimal k value for all training samples by developing new sparse reconstruction model. A sparse reconstruction model essentially contains the correlation coefficients i.e. the weights that are estimated for the travel-time (for the five days) and these correlation coefficients form the base for constructing the weight matrix also called as the k matrix. From the k matrix, the k value is obtained which are essentially the non-zero values in the k matrix. In the test stage, the k-Tree fast outputs the optimal k value for each test sample, and then, the k-NN prediction can be conducted using the learned optimal k value. The developed k-NN algorithm was tested on a study corridor of 59.43 km in the Mumbai arterial roads using public transport (bus) as the study mode. The standard and proposed k-NN models gave reasonable MAPE values (average value obtained for five days) as 20.43 and 10.15 respectively. The proposed k-NN method gives MAPE value of about 10% indicating that by optimizing the k value and by training the data; there is a higher chance for the k-NN model to predict accurately.

  • Supplemental Notes:
    • This paper was sponsored by TRB committee ABJ70 Standing Committee on Artificial Intelligence and Advanced Computing Applications.
  • Corporate Authors:

    Transportation Research Board

  • Authors:
    • Sharmila, R B
    • Velaga, Nagendra R
  • Conference:
  • Date: 2019


  • English

Media Info

  • Media Type: Digital/other
  • Features: Figures; References; Tables;
  • Pagination: 6p

Subject/Index Terms

Filing Info

  • Accession Number: 01697463
  • Record Type: Publication
  • Report/Paper Numbers: 19-04359
  • Files: TRIS, TRB, ATRI
  • Created Date: Mar 1 2019 3:50PM