Real-Time Grain Commodities Price Predictions in South Africa: A Big Data and Neural Networks Approach. Issue 4 (1st October 2016)

• Record Type: Journal Article
• Authors: Ayankoya, Kayode; Calitz, Andre P.; Greyling, Jean H.
• Is Part Of: Agrekon. Volume 55:Issue 4(2016)
• Page Start: 483
• Page End: 508
• Publication Date: 2016-10-01
• Subjects: Big Data -- Neural Networks -- grain commodities prices -- predictions -- white maize; Q11 -- Q13 -- C63 -- C89; Agriculture -- Economic aspects -- Africa, Southern -- Periodicals; 338.10968; 338.10968 A277
• DOI: 10.1080/03031853.2016.1243060
• Languages: English
• ISSNs: 0303-1853

Abstract

ABSTRACT: The prices of agricultural grain commodities are known to be volatile due to several factors that influence these prices. Moreover, different combinations of these factors, such as demand, supply and macroeconomic indicators are responsible for the price volatility at different times. Big Data presents opportunities to collect and integrate datasets from several sources for the purpose of discovering useful patterns and extracting actionable insights that can be used to gain competitive advantage or improve decision making. Neural Networks presents research opportunities for training computer algorithms to model linear and non-linear patterns that might exist in datasets for the purpose of extracting actionable insights such as making predictions. This article proposes a Big Data and Neural Networks approach for predicting prices of grain commodities in South Africa. It was identified that disparate data that influence the grain commodities market can be acquired, integrated and analysed in real-time to predict future prices of grain commodities. By utilising SAP HANA as the enabling Big Data technology, data acquired from several sources was used to create an integrated dataset, and a predictive model was developed using Backpropagation Neural Network algorithms. This model was used to predict the daily spot prices of white maize on the Johannesburg Stock Exchange (JSE) at the end of each trading day. The initial results indicate that the approach can be scientifically used to predict future prices of grain commodities in a real-time environment.