ARBA MINCH UNIVERSITY ARBA MINCH INSTITUTE OF TECHNOLOGY FACULTY OF COMPUTING AND SOFTWARE ENGINEERING

Abstract

country’s economy including agriculture and manufacturing industries. Since Ethiopia is a developing country, the market that take place in this country is running in traditional and also the analysis is done in fundamental and statistical principle. The collected data by governmental and non-governmental organization through different mechanisms are not used to forecasting future market direction in Ethiopia. As a result, farmers, investors, traders, business sectors and commodity exchange organizations in the country are challenged to high business risk and terrified to invest on agricultural sector. This research study aims to build an effective commodity market analysis predictive model for Ethiopian commodity exchange. The study examines attributes of historical Ethiopian Market Data (EMD) to find the most potential attributes for forecasting future market direction and price. To select the potential attributes from datasets, various attributes of EMD are tested for individual forecasting ability. The researcher found 9 most potential attributes for Ethiopian market price forecasting. The study also examines four supervised machine learning (SVL) and one hybrid algorithms for market price forecasting on historical EMD. Different models are developed using a separate training and testing and a 10-fold cross validation test options with four datasets using attributes with high forecasting ability and low redundancy level. Frommodels, Multilayer Perceptron (MLP), Random Forest (RF), M5rules, Support Vector Machine (SVM) and Hybrid algorithm (SVM and MLP), the forecasting accuracy of SVM and Hybrid algorithms are shown to be very accurate with minimum forecasting error and maximum correlation factor than M5rules, MLP and RF. SVM attained the highest forecasting accuracy with Mean Absolute Error (MAE) of 0.8104, Root Mean Square error (RMSE) of 1.91975 and Coefficient Correlation (CC) of 0.99725. And the hybrid algorithm also attained better accuracy than others with MAE of 1.629475, RMSE of 2.7148 and of 0.99625, whereas RF attained the worst forecasting performance to forecast market price of Ethiopian commodities with high MAE of 6.71055, RMSE of 14.433475 and low CC value of 0.95275. Based on this research study finding, it is possible and reasonable to conclude that forecasting Ethiopian market price using SVL and hybrid algorithms were appropriate and satisfactory to real application.