Rangkaian Neural dalam Peramalan Harga Minyak Kelapa Sawit

Abstract

Kertas kerja ini membincangkan penggunaan rangkaian neural suap-kehadapan dengan satu aras tersembunyi digabungkan dengan algoritma rambatan balik dan didapati ia sesuai untuk memerihalkan data harga minyak sawit. Kajian awal yang telah dilakukan oleh Azme et al. mendapati analisis regresi berganda kurang sesuai digunakan kerana masalah multikolineariti dalam data kajian. Lima harga minyak sayuran dunia iaitu minyak sawit mentah, minyak isirong, minyak kacang soya, minyak kelapa dan minyak biji sawi telah dianalisis. Dua model telah dicadangkan iaitu, NN1 dan NN2. Hasil kajian mendapati bahawa kedua-dua model telah menunjukkan prestasi yang tinggi dengan mencatatkan nilai pekali penentuan, R2 yang tinggi iaitu 0.938 dan 0.940 masing-masing. Umumnya, rangkaian neural berupaya menjadi satu kaedah alternatif sekiranya masalah multikolineariti wujud terhadap data yang dikaji. Kata kunci: Harga minyak sawit, multikolineariti, rangkaian neural suap-kehadapan, fungsi penggiat, algoritma rambatan balik The application of feed forward neural network with one hidden layer, which is combined with back propagation algorithm, has been discussed in this paper. This study found that the combination of these two methods was reliable to explain the palm oil price data. A research was done earlier by Azme et al. and found that multicollinearity problem occurred in data analysis when multiple linear regression was applied. We considered five vegetable oil prices namely, crude palm oil, soybean oil, palm kernel oil, coconut oil and rapeseed oil in this study. Two models have been proposed, i.e. NN1 and NN2 and these models performed very well with the determination coefficient R2, were 0.938 and 0.940 respectively. Generally, neural network is able to be an alternative method if multicollinearity problem occurred in the surveyed data. Key words: Palm oil juice, multicollinearity, feed forward neural network, activation function, back propagation algorithm