Stock Market LSTM Forecasting

LSTM stands for Long Short Term Memory Networks. Many theories suggest stock prices tend to depend on their previous values, despite many assumptions of a random walk model. In this project, I attempted to study the impact of trend of a stock on its future prices.

I took a lag of 4 prices, which means that I judged the dependance of a stock's price today on four previous closing values. I tried to build a relationship for the same and aimed to predict the coeffecients accurately.

Preprocessing

Used MinMax scaler to compress the stock price data between 0 and 1 as LSTM is very sensitive to the scale of the data. Then, divided the data into training and testing batches, in a ratio of 2:1. Cross Validation and Random Seed work better in Regression so not done here.

Timesteps : talks about how many previous days does present day price depend on

Previous days are X-Train and Today is Y-Train.

Training the LSTM Model

Imported Sequential, Dense and LSTM libraries. Ran 100 epochs. Scaler inverse transform was done on previously transformed data to study RMSE easily.

Testing the model over the test dataset

Predicting output for the next 30 days