Multivariate Time Series Forecasting Based Cloud Computing For Consumer Price Index Using Deep Learning Algorithms

Abstract

Multivariate time series forecasting affords an opportunity to forecast future recent trends or possibility incident based on historical observations. Forecasting in economic world becomes global interest particularly for researchers seeking for best accuracy result using several methods. Consumer Price Index is the primary instrument used by central banks to set inflation targets. However, most of previous studies commonly only used univariate factor to forecast Consumer Price Index. Furthermore, mostly model development of forecasting system is done by personal and physical server facing the problem of impractical yet time consuming. Since measuring method of Consumer Price Index commonly is pick an average of the period-to-period price move for the different products, we conducted multivariate Consumer Price Index forecasting based Cloud Computing utilizing 28 types of Surabaya daily food price from 2014 to 2018 using Multilayer Perceptron and Long Short Term Memory (LSTM) of deep learning. Furthermore, we implement architectural variations of the number of neurons, epoch, and hidden layers. The whole development of forecasting system is built in Amazon Web Service (AWS) Cloud. The result indicated the best accuracy value was obtained from the Multilayer Perceptron with 3.380 of RMSE consist of a configuration of 2 hidden layers, 10 neurons of first hidden layer, 10 neurons of second hidden layer also 1000 of epoch.