Analisis peramalan kausal berbasis integrasi principal component analysis dan jaringan saraf tiruan untuk aplikasi teknik industri

SUTONO, Sugoro Bhakti, Ir. Subagyo, Ph.D

2008 | Tesis | S2 Teknik Mesin

Penelitian ini mengembangkan model peramalan kausal berbasis integrasi dari principal component analysis dan jaringan saraf tiruan. Tujuan penelitian ini adalah untuk mengetahui pengaruh penerapan principal component analysis terhadap tingkat akurasi hasil peramalan berbasis jaringan saraf tiruan. Dalam penelitian ini, analisis akurasi metode yang dikembangkan dilakukan dengan diuji menggunakan 16 set data kasus peramalan yang terdiri dari satu variabel terikat dan beberapa variabel bebas. Tahap penelitian meliputi pemilihan dan analisis variabel kausal, tahap principal component analysis, tahap peramalan dengan jaringan saraf tiruan, tahap analisis hasil peramalan. Analisis tingkat akurasi peramalan dilakukan berdasarkan parameter error MAPE. Pemilihan dan analisis variabel kausal dilakukan berdasarkan uji korelasi dan uji hipotesis student t-test. Sementara penentuan jumlah principal components dilakukan dengan prosedur proportion of trace explained. Sedangkan model jaringan saraf tiruan yang dibangun adalah jaringan saraf tiruan propagasi-balik arsitektur satu hidden layer dengan fungsi pelatihan metode resilient backpropagation. Hasil penelitian menunjukkan bahwa perlakuan principal component analysis berpengaruh terhadap peningkatan tingkat akurasi peramalan menggunakan metode jaringan saraf tiruan. Penerapan principal component analysis secara keseluruhan memberikan pengaruh penurunan rata-rata nilai MAPE sebesar 35,53% dibandingkan rata-rata nilai MAPE hasil peramalan jaringan saraf tiruan. Berdasarkan analisis peta kontrol nilai tracking signal menunjukkan bahwa principal component analysis cenderung menghilangkan bias error peramalan atau signal tripped metode yang dikembangkan.

The research develops causal forecasting model based on the integration of principal component analysis and artificial neural networks. It aims at identifying the effects of principal component analysis application on the degree of accuracy of the forecast outcome which is based on artificial neural networks. In this research, the method developed for the accuracy analysis employs a test using 16 data sets of forecast cases which comprise one dependent variable and a number of independent variables. Stages incorporated in the research are causal variable selection and analysis stage, principal component analysis stage, forecasting stage using artificial neural networks, and analysis of forecast outcome stage. The degree of accuracy of the forecast outcome is measured in MAPE error parameter. Causal variable selection and analysis is based on correlation test and student’s t-test hypothesis test. The number of principal components is determined by following the procedure for proportion of trace explained. The model of artificial neural networks constructed is one-hidden layer backpropagation artificial neural networks architecture with resilient backpropagation training function method. The results of the research indicate that the treatment of principal component analysis has effects on raising the degree of accuracy in the forecast using artificial neural networks. Overall, the application of principal component analysis decreases the average value of MAPE to the extent of 35,53% compared to that of artificial neural networks forecast outcome. According to the control chart analysis of tracking signal value, the results show that principal component analysis tends to leave out bias of the forecast errors or signal tripped of the developed method.

Kata Kunci : Peramalan kusal,Principal component analysis,Jaringan saraf tiruan,causal forecasting, data transformation, principal component analysis, artificial neural networks, accuracy