Stock Price Prediction of Several Banks Using LSTM Recurrent Neural Network

Dataset :

The dataset is taken from library yahoo_historical. That consists of Open, High, Low, Closing and Volume of several banks (Bank BCA, Bank BRI, Bank BNI and Bank Mandiri) from 12th July 2012 until 12th July 2019 with total 1739 rows in each stock. Variabel that used for prediction is Open Price.

from yahoo_historical import Fetcher
namasaham = "bbca"
namasaham = namasaham.upper()
saham = Fetcher(namasaham+".JK", [2012,7,12], [2019,7,12], interval="1d")
saham = saham.getHistorical()
# Open Price
saham = saham.iloc[:,1:2]  

Data Pre-Processing :

1. Remove rows with NA values in Open Price

   saham = saham.dropna()  

2. Describe data

   • BBCA

          count          mean          std     min      25%      50%      75%      max
   Open  1739.0  15829.700978  5896.481231  7450.0  11025.0  13625.0  20887.5  30100.0

   • BBRI

          count         mean         std     min     25%     50%     75%     max
   Open  1739.0  2476.138585  778.166236  1270.0  1927.5  2300.0  3025.0  4510.0

   • BBNI

          count         mean         std     min     25%     50%     75%     max
   Open  1739.0  2476.138585  778.166236  1270.0  1927.5  2300.0  3025.0  4510.0

   • BMRI

          count         mean         std     min     25%     50%     75%     max
   Open  1739.0  2476.138585  778.166236  1270.0  1927.5  2300.0  3025.0  4510.0

3. Split dataset into data training and data testing with train size=75% and test size=25 %

   percent_train = 0.75
   trainingset = saham.iloc[:int(len(saham)*percent_train),:]
   testset = saham.iloc[int(len(saham)*percent_train):,:]

4. Normalized data between zero and one with MinMaxScaler

   from sklearn.preprocessing import MinMaxScaler
   
   sc = MinMaxScaler()
   trainingset = sc.fit_transform(trainingset)       
   testset = sc.fit_transform(testset)

5. Build trainX, trainY, testX and testY (with window = 3)

   def createDataset(data, window):
       dataX, dataY = [], []
       for i in range(len(data)-window):
           temp = []
           for j in range(i, i+window):
               temp.append(data[j,0])
           dataX.append(temp)
           dataY.append(data[i+window,0])
       return np.array(dataX), np.array(dataY)   

Model :

• Dropout = 0.2
• Epochs = 500
• LSTM 50

  # Model Sequential
  regressor=Sequential()
  # Hidden layer (4)
  regressor.add(LSTM(units=50,activation='tanh',return_sequences=True, input_shape=(trainX.shape[1],1)))
  regressor.add(Dropout(0.2))
  
  regressor.add(LSTM(units=50, return_sequences=True))
  regressor.add(Dropout(0.2))
  
  regressor.add(LSTM(units=50, return_sequences=True))
  regressor.add(Dropout(0.2))
  
  regressor.add(LSTM(units=50))
  regressor.add(Dropout(0.2))
  
  # Output layer (1)
  regressor.add(Dense(units = 1))
  
  regressor.compile(optimizer='adam',loss='mean_squared_error')
  regressor.fit(trainX,trainY,epochs=500)

Predict data testing :

predict_price = regressor.predict(testX)

Denormalization data for accuration and plotting :

testY = sc.inverse_transform(testY.reshape(-1,1))
predict_price = sc.inverse_transform(predict_price)
trainingset = sc.inverse_transform(trainingset)
testset = sc.inverse_transform(testset)

Accuracy of model prediction :

with Directional Statistics (Dstat) that more accurate commonly used in special financial economic fields.

def dstat(x,y):
    dstat = 0
    n = len(y)
    for i in range(n-1):
        if ((x[i+1]-y[i])*(y[i+1]-y[i]))>0 :
            dstat += 1
            Dstat = (float(1/(n-1))*dstat)*100
    return float(Dstat)

Ploting the result of data testing prediction :

# plot real data testing
plt.plot(np.arange(len(testY)),testY,color='red',label='Real stock price of {}'.format(namasaham))

# plot predict data testing
plt.plot(np.arange(len(predict_price)),predict_price,color='blue',label='Predicted stock price of {}'.format(namasaham))

plt.title('Predicting Data Testing Stock Price of {}\nScore Dstat ={}%'.format(namasaham,round(dstat(testY,predict_price),3)))
plt.xlabel('Time')
plt.ylabel('Stock price')
plt.legend()
plt.savefig('/templates/testing_predict_{}.png'.format(namasaham))
plt.show()