Abstract
Using MATLAB’s Perceptron model, this paper presents an attempt to train a neural network distinguish between acceptable and unacceptable purchases of publicly traded stock. In the past, Perceptron models have been used, quite successfully, in similar classification exercises. The input vectors used in training the network and in making the classifications in our model, involve readily available financial data like current ratio; quick ratio ; gross margin as a percentage; sales/asset turnover ; and earnings per share. The initial results of our analysis were quite encouraging insofar as the model had a ninety percent prediction accuracy using held-back test data. On the basis of our initial success , we are currently trying to extend this model to a "forward-looking" investment decision process model.
Recommended Citation
Roy, Probir, "A "Go/No' Perceptron Stock Classification Model" (1998). AMCIS 1998 Proceedings. 72.
https://aisel.aisnet.org/amcis1998/72