RNN for Time series Forecasting Using Google Stock Prices

Abstract

This thesis investigates the use of Long Short-Term Memory (LSTM) networks to predict Google’s stock prices. The study focuses on stock data from January 2012 to December 2016 for training, and January 2017 for testing. LSTM, a type of Recurrent Neural Network (RNN), is ideal for time series forecasting because it can learn long-term dependencies. To prepare the data, stock prices were normalized using Min Max Scaler, which helps improve model performance. The data was then organized into sequences of 60-time steps using a sliding window approach. The LSTM model was built with four layers, each containing 50 units, and included dropout layers to reduce overfitting. Training was conducted using the Adam optimizer and mean squared error as the loss function over 100 epochs with a batch size of 32. The results showed that the LSTMmodel effectively captured the stock price patterns, highlighting its potential for accurate financial forecasting.