A Randomized Algorithm for Prediction Interval Using RVFL Networks Ensemble

Prediction Intervals (PIs) can specify the level of uncertainty related to point-based prediction. Most Neural Network (NN)-based approaches for constructing PIs suffer from computational expense and some restrictive assumptions on data distribution. This paper develops a randomized algorithm for PIs building with good performance in terms of both effectiveness and efficiency. To achieve this goal, a neural network ensemble with random weights is employed as a learner model, and a novel algorithm for generating teacher signals is proposed. Our proposed Randomized Algorithm for Prediction Intervals (RAPI) constructs an NN ensemble with two outputs, representing the lower and upper bounds of PIs, respectively. Experimental results with comparisons over nine benchmark datasets indicate that RAPI performs favourably in terms of coverage rate, specificity and efficiency.