Instance selection is a way to remove unnecessary data that can adversely affect the prediction model, thereby selecting representative and relevant data from the original data set that is expected to improve predictive performance. Instance selection plays an important role in improving the scalability of data mining algorithms and has also proven to be successful over a wide range of classification problems. However, instance selection using an evolutionary approach, as proposed in this study, is different from previous methods that have focused on improving accuracy performance in the stock market (i.e., Up or Down forecast). In fact, we propose a new approach to instance selection that uses genetic algorithms (GAs) to define a set of target labels that can identify the buying and selling signals and then select instances according to three performance measures of the trading system (i.e., the winning ratio, the payoff ratio, and the profit factor). An intelligent ensemble trading system with instance selection using GAs is then developed for investors in the stock market. An empirical study of the proposed model is conducted using 35 companies from the Dow Jones Industrial Average, the New York Stock Exchange, and the Nasdaq Stock Market from January, 2006 to December, 2016.