Table 5 BACKPROPAGATION ACTUAL AND PREDICTED STOCK PRICES (SNAPSHOT)
Related Figures (12)
Fig. 2. Time Window For Lucent Inc. Fig. 1. Time Window For SouthWest Airlines BACKPROPAGATION MODEL EXPERIMENTS Eventually the chosen network went through a change in learning rate to determine if the network can learn at either a faster, medium or slower rate. As shown in the bottom of Table I, the change in learning rate does not produce the least difference between the training and testing data. The negative difference in Table I depicts that the network might have a high probability of classification errors. TABLE I KOHONEN SOM NETWORK MODEL EXPERIMENTS Network Architecture—Kohonen SOM: Several Kohonen SOM models were experimented with. The best model was chosen from Table II again, in bold. The network has 64 inputs with 32 Kohonen neurons and | output. HYBRID KOHONEN SOM NETWORK MODEL EXPERIMENTS CLASSIFICATION PERFORMANCE OF THE NETWORK and finally into an output layer just as shown in Figure 4. In other words, the Kohonen SOM is the front-end, while the hidden and output layer of a MLP is the back-end of the Hybrid Kohonen SOM. Generally, the Kohonen SOM network is not specified by input, hidden and output layers. However, the input, hidden and output layers were included in figure 4 for illustration purposes. Since the research objective is to predict the stock price of the day, Figure 5 shows that BPN network model predicts inaccurate stock prices when the price is higher than $60. Fig. 5. Backpropagation Model—Actual Vs. Predicted Prices Fig. 6. Kohonen SOM Model—Actual Vs. Predicted Prices Fig. 7. Hybrid Kohonen SOM Model—Actual Vs. Predicted Prices KOHONEN SOM ACTUAL AND PREDICTED STOCK PRICES (SNAPSHOT)
