Paper Title
A Case Study of Stock Investment Based on Neural Network Techniques

Abstract
Stock price predictions based on data mining techniques have been widely examined. Most studies focused on predicting the stock price or up/down in the next trading day. Considering a future time horizon (say, 100 days), this research attempts to predict whether a day is a “buy-day”. A buy-day denotes a “good” day to purchase a particular stock if the stock closing price is expected to rise over 10% in the coming 100 days. The “buy-day” decision is a binary classification problem, which herein shall be solved by various artificial neural network (ANN) models. In an ANN model, the output involves two classification states: “buy-day” or “not-buy-day”; and the input can involve up to 15 variables (i.e., features). By selecting different portfolios of input features, three ANNs are established. The stock price ranging from Jan. 2007 to Dec. 2016 (10 years) of a Taiwanese company (Foxconn) is used as a test case. Numerical experiments reveal that the 3rd ANN outperforms the other ANNs; and its average annual return of investment is about 10.92%. Keywords - Binary classification, Data mining, Neural network, Stock price prediction