FORECASTING SOFTWARE MARKET PRICE USING BACK PROPAGATION NEURAL NETWORK

Abstract

This research uses backpropagation artificial neural network to examine whether it is capable of adequately capturing software cost complexities in weight space, to enable it to make accurate estimates.The input for this task is a set of open information about the software of a certain type.Because of the openness criterion, it was decided to choose mobile apps for Android, full information of this type of software is available on an open source Google Play Market.The purpose of the developed software is to reach the Google server in real time and to receive up-to-date information on the current situation in the mobile application market.Within the collected data set, backpropagation artificial neural network appears to indicate the potential to be developed into good software price estimation models.The model is not difficult to develop and has the flexibility of being able to incorporate additional attributes as input if special circumstances warrant their inclusion.Neural network has the ability to capture knowledge of the complex interrelationships in weight matrix to make predictions.For this research the data were divided into three sets.The training set, the test set, and the validation set.The data for each category were randomly chosen, except that the data in the test and validation sets was not allowed to be larger or smaller than the largest and smallest features respectively in the training set.This was done so that predictions were not made outside the data range on which the network had been trained.The inputs were rating, number of ratings, number of downloads, number of reviews, in-app purchases, number of supported languages.The accuracy of the price estimate was taken as the Root Mean Square Error (RMSE).