Abstract :
Neural Network Algorithm is one of the supervised learning algorithms with a high degree of accuracy in forecasting, but the Neural Network still has shortcomings in determining the maximum weight value so that it needs to be optimized. Therefore, an optimization method is needed to cover the shortcomings of the Neural Network algorithm. There are several optimization methods such as Genetic Algorithms, Particle Swam Optimization (PSO), Bootstrap Aggregating (Bagging), Adaptive Boosting (AdaBoost) and so on. This study using the Bootstrap Aggregating (Bagging) method. The Bagging method was chosen because it made small changes during training but it can change the results on the model used so that it is considered to be able to cover the shortcomings of the Neural Network algorithm to determine the maximum value weight. This research was conducted on the JCI data for the last 15 years from 2005-2020 which were taken from Yahoo Finance. The data processing in this study starts from preprocessing to determine attributes, eliminating null data, normalizing and windowing processes then dividing the dataset using 10k-fold validation into training data and testing data. After that, search for each best parameter through the predetermined initial parameters. Neural Network algorithm process and optimization using the Bagging method with predetermined initial parameters and also the best parameters that have been obtained. The results of this study indicate that the average RMSE value is better when optimized using the Bagging method, which is 46.22179 with the best parameters in the Neural Network algorithm and the best parameters in the Bagging method. While the average RMSE value of the Neural Network algorithm before being optimized both in the initial and best parameters is 49.14626 and 48.65204. This shows that the Bagging method can optimize the Neural Network algorithm on the JCI data for the last 15 years. Keyword : Bagging, IHSG, Neural Network, Optimalisasi, RMSE