Chapter

Neural Information Processing

Volume 5864 of the series Lecture Notes in Computer Science pp 141-149

Improving Prediction Interval Quality: A Genetic Algorithm-Based Method Applied to Neural Networks

  • Abbas KhosraviAffiliated withCentre for Intelligent Systems Research (CISR), Deakin University
  • , Saeid NahavandiAffiliated withCentre for Intelligent Systems Research (CISR), Deakin University
  • , Doug CreightonAffiliated withCentre for Intelligent Systems Research (CISR), Deakin University

* Final gross prices may vary according to local VAT.

Get Access

Abstract

The delta technique has been proposed in literature for constructing prediction intervals for targets estimated by neural networks. Quality of constructed prediction intervals using this technique highly depends on neural network characteristics. Unfortunately, literature is void of information about how these dependences can be managed in order to optimize prediction intervals. This study attempts to optimize length and coverage probability of prediction intervals through modifying structure and parameters of the underlying neural networks. In an evolutionary optimization, genetic algorithm is applied for finding the optimal values of network size and training hyper-parameters. The applicability and efficiency of the proposed optimization technique is examined and demonstrated using a real case study. It is shown that application of the proposed optimization technique significantly improves quality of constructed prediction intervals in term of length and coverage probability.

Keywords

Neural network genetic algorithm prediction interval