TECHNIQUES TO OPTIMIZE DATA COLLECTION AND TRAINING TIME FOR FORECASTING STOCK MARKET TRENDS USING ARTIFICIAL NEURAL NETWORK

摘要

The data required to train Neural-Net works for trading systems typically consist of historical price databases regarding price shares connected to decision-making process of a particular company. An important point to note is that the data to be collected must have low noise or errors which would otherwise cause the production of inaccurate solutions/forecasts. The training process can last from hours to weeks, depending on the amount of data that is used for the training. This paper focuses on the data collection and input selection based on the relative contribution of selected technical indicators. More specifically, the paper focuses on preprocessing the available data in terms of moving averages for specified time periods and observing the strength of prediction and thereafter selecting a final set of inputs.