부산대학교 도서관

Present artificial intelligence advances tend to be focused on customized deep learning techniques which are computational expensive and require costly infrastructure. These techniques have shown to be particularly effective in highly complex environments such as image processing, natural language processing and market price predictions. On the other hand, small companies are requiring more and more access to artificial intelligence to predict customer behavior and hence to avoid to be affected by the highly volatility and variance of the market. Unfortunately, most of these companies may not be able to afford the costs of current artificial intelligence advanced methods. Hence, in this paper we study a low-cost known alternative: decision tree classifiers. In particular, we focus our analysis on the benefits to use them to analyze market predictions with high area under the receiver operating characteristic curve over three databases: Social Network Advertising Sells, Organic Purchased Indicator, and Online Shoppers Purchasing Intention. The best decision tree models obtained were those that produced an area under the receiver operating characteristic curve score from 0.81 to 0.96. In addition, we report the accuracy of our models which provided results ranging from 79.80& to 89.80&. These results show that simple models like decision trees are good to understand the fluctuation and trends from market data, and since its simplicity are an alternative for small businesses willing to try artificial intelligence predictions.