Apple and google stock prediction using recurrent networks
Every stock data has 4 important data:
Open: the price at the beginning of the sale that day Close: the last price that it was closed for sale that day Low: the lowest price it has reached that day High:the highest price it has reached that day
Here I used the 'Close' price to predict the price for the other days. Another more accurate way is to use the average of the Low and High price.
This is what happened in this code:
-
making a time-series of data. Every day data is made by the last 3 days data.
-
split it in Train, Validation and Test sets by Fold-sampling
-
The model details tested with three recurrent cells : RNN , LSTM, GRU tested with 3 optimizers: Adam, RMSprob, AdaDelta tested with learning-rate of 0.001 250 epochs
the models reached 99% accuracy and the lowest misprediction rate was: MAPE: 1.15 for Google MAPE: 3.00 for Apple