Web Browser-Based Forecasting of Economic Time-Series

Time series forecasting using evolutionary neural nets implemented in a volunteer computing system

“…In order to test the capabilities of web browsers in financial time‐series forecasting, two experiments have been carried out using jsEvRBF , updating those performed in our previous work (Rivas et al, ). For the first experiment (see section 4.1) and in order to test the performance in a volunteer computing set up, we have effectively opened the system to a wide community of users, so that many different devices, using different operating systems and web browsers have participated in it.…”

“…With respect to the accuracy of the forecasting performed in each experiment, the MSE of the best solution found along all the executions was 6×10 −4 . As can be see in Table , this solution is as good as the average one found by the original algorithm (EvRBF) and much better that our previously published results (Rivas et al, ). The later is due to the fact that most of the executions have finished with valid values, since new versions of Safari and Chrome browsers (i.e., AppleWebKit‐based browsers) seems to be less affected by the NaN error we found in our previous work.…”

“…A second experiment has been done attempting to detect differences between the different browsers currently available. Existing errors previously reported in our paper for PAAMS 2016 (Rivas et al, ) have disappeared in browser Safari, but still remains in Chrome. For this reason, although Chrome is a good candidate to run the algorithm, should be rejected in favor of Firefox.…”

“…Regarding the herein described implementation, our research group has already published the preliminary results obtained by the initial proof of concept in (Rivas, Parras‐Gutierrez, Merelo, Arenas, & García‐Fernández, ). The results presented in this new paper correspond to new experiments that measure the performance obtained with different browsers and in different conditions.…”

“…The results presented in this new paper correspond to new experiments that measure the performance obtained with different browsers and in different conditions. In order to compare results, jsEvRBF has been evaluated with the same time‐series used by Sheta and de Jong (Sheta & de Jong, ) and the results were published in (Rivas et al, ). For this paper, we go beyond the state of the art by including new experiments and new results.…”