ABSTRAKSI: Tujuan utama tugas akhir ini adalah mengimplementasikan dan menganalisa bagaimana Evolutionary Programming dapat digunakan untuk mengoptimasi bobot jaringan saraf tiruan Menganalisa bagaimana pengaruh pemilihan parameter – parameter Evolutionary Programming yang berupa jumlah generasi, ukuran populasi dan operator mutasi terhadap bobot hasil keluaran jaringan saraf tiruan. Implementasi optimasi jaringan saraf tiruan dengan Evolurionary Programming diterapkan pada sistem peramalan harga saham. Arsitektur jaringan saraf tiruan yang dibangun adalah feed forward neural network (FFNN) yang bersifat statis, 5 neuron pada input layer, 3 neuron pada hidden layer dan 1 neuron pada output layer. Inputan sistem berupa data historis harga saham untuk analisa teknikal yaitu harga penutupan (close), harga tertinggi (high), harga terendah (low), volume transaksi dan index IHSG, sedangkan keluaran sistem berupa hasil prediksi harga penutupan (close) satu hari ke depan. Pada penelitian tugas akhir ini didapat kesimpulan bahwa Sistem peramalan harga saham yang dibangun dengan Jaringan Syaraf Tiruan dengan metode pelatihan Evolutionary Programming dengan Arsitektur jaringan [5 3 1], fungsi aktivasi logsig dan purelin, range bobot [-1.5 1.5], range mutation step [0.01 0.1], operator mutasi Gaussian 0.15 dan 0.10, normalisasi dengan 2 standar deviasi dan data terurut, populasi 50 , generasi 1000, dan jumlah lawan q adalah 10% ukuran populasi, ini mampu memberikan solusi bobot yang terbaik dimana dihasilkan hasil prediksi dengan error yang paling kecil untuk data training dan data testing masing-masing dalam Mean Absolute Percentage Error (MAPE) yaitu 1.79 % dan 2.38%
Kata Kunci : Evolutionary programming, jaringan syaraf tiruan, mutasi, standar deviasi, mean absolute percentage error, analisa teknikal sahamABSTRACT: The main goal of this final project is to implement and analyze how Evolutionary Programming is applicable to optimize the weight of artificial neural network and Analyze the influence of the election of Evolutionary Programming parameters such as number of generation, number of population and mutation operator to the result of artificial neural network output. implementation of optimize artificial neural network by Evolutionary Programming applied at share price forecasting system. Artificial neural network architecture that build is feed forward neural network (FFNN) with 5 neuron at input layer, 3 neuron at hidden layer and 1 neuron at output layer. The input for this system is share price historical data which is usually used for technical analyze that is closing price ( close), highest price ( high), lowest price ( low), volume of transaction and index of IHSG, while the system output is the prediction of closing price ( close) one day forwards. At this final project research, got a conclusion that share price forecasting system that build with the artificial neural network using evolutionary programming as a training method with network architecture is [ 5 3 1], activation function is logsig and purelin, weight range is [-1.5 1.5], mutation step range is [ 0.01 0.1], the Gaussian mutation operator is 0.15 and 0.10, normalization by 2 standard deviation and sequence data, population 50, generation 1000, and number of the opponent q is 10% of size population, able to give the best weight solution where the prediction result error is smallest for data training and data testing in Mean Absolute Percentage Error ( MAPE) that is 1.79 % and 2.38%.
Keyword: Evolutionary Programming, artificial neural network, mutation, standard deviation, mean absolute percentage error, share price technical analyze