Ethiopian Agricultural Market Price Data Analysis and Prediction using Deep Learning Models

Ahmed, Kebede

dc.contributor.author	Ahmed, Kebede
dc.date.accessioned	2021-08-13T06:56:59Z
dc.date.available	2021-08-13T06:56:59Z
dc.date.issued	2021-03
dc.identifier.uri	http://ir.bdu.edu.et/handle/123456789/12396
dc.description.abstract	Ethiopia Commodity Exchange (ECX) is a commodity exchange established April 2008 and it is a national multi-commodity exchange that brings together buyers and sellers of agricultural commodities. It has been collecting a handful of time series data in each year. An accurate and efficient prediction of such time series data is conducive for evading market risks and supports the agricultural system that realize the balance of supply and demand of agricultural commodities. However, the agricultural commodity market is traditional and the market price data is not used for dynamic market price data analysis and prediction. Many countries proposed and utilizing different prediction mechanism for the future price of their market. However, the research approaches are different and the computational algorithms also based on the real market situation. Moreover, the market features that have impact on one country may not have similar impact in another country. Hence, the intention of this thesis work is analyzing and predicting the target market using deep learning models based on both fundamental and technical market analysis approaches. Four most commonly used deep learning models were explored for their predictive ability of target market price based on both the fundamental and technical indicator features of coffee, sesame and haricot bean and wavelet transform was used for denoising the price data. The technical indicators calculated from price data were given to the Boruta feature selection algorithm and 39 features for Coffee, 24 for Sesame and 27 features for Haricot bean were selected. The sentiment scores were calculated from the financial news related to each commodity. Lastly, the denoised price data together with selected technical indicator and sentiment scores were used as an input to the models. Three experiments were conducted for each commodity; the first was model building and prediction with only the price data. The second was using the price data with commodity technical indicators (CTI) and the third one was using price data, CTI and sentiment scores. From models, (Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM), convolutional neural network (CNN) and CNN-LSTM network), the performance of LSTM is shown to be the best fit model with minimum Mean Squared Error (MSE) of 9.6x10-5., which is small enough, and it indicates that the predictive model performs well.	en_US
dc.language.iso	en_US	en_US
dc.subject	computer science	en_US
dc.title	Ethiopian Agricultural Market Price Data Analysis and Prediction using Deep Learning Models	en_US
dc.type	Thesis	en_US