Stock price prediction using Neural Net
12 likes11,576 views
The document discusses stock price prediction, emphasizing the roles of various machine learning and deep learning models in forecasting stock prices. It details the selection of relevant features and compares linear regression with recurrent neural networks, specifically LSTM, highlighting the strengths of LSTM in avoiding long-term dependency issues. The findings suggest that deep learning models outperform traditional machine learning models in predicting stock prices.
Stock price prediction using Neural Net
- 2. INTRODUCTION ● A stock market is a public market for the trading of company stock. ● Stock market allows us to buy and sell units of stocks (ownership) of a company. ● If the company’s profits go up,then we own some of the profits and if they go down, then we lose profits with them. ● If more sellers than buyers, stock prices tend to fall. Conversely, when more buyers than sellers, stock prices tend to rise
- 3. INTRODUCTION ProfitLoss Red is purchased stock Green is sold stock
- 4. HOW TO READ A STOCK TABLE? Date- day on which the stock is traded High- high is the highest price at which a stock traded during the course of the day Close- refers to the last price at which a stock trades during a regular trading session Open- price of the first trade for any listed stock is its daily opening price. Low- lowest price at which a stock trades over the course of a trading day. Volume- the number of shares or contracts traded in a security or an entire market during a given period of time
- 5. ● To accurately predict the future closing value of a given stock across a given period of time in the future. ● Use different machine learning and deep learning models available and compare them in terms of graphical analysis. 5 PROBLEM STATEMENT Actual Close Predicted Close Today’s Close price Predicted Close price of Next day
- 6. The data used in this project is of the Alphabet Inc3 from January 1, 2005 to July 30, 2017, this is a series of data points indexed in time order or a time series. Our goal was to predict the closing price for any given date after training 6 DATASET
- 7. ● Process of selecting a subset of relevant features for use in model construction. ● Feature selection methods include and exclude attributes present in the data without changing them. ● Here, in our case ‘Date’,‘High’ and ‘Low’ attributes are dropped. 7 FEATURE IMPORTANCE Dropped
- 8. ● Process of selecting a subset of relevant features for use in model construction. ● Feature selection methods include and exclude attributes present in the data without changing them. ● Here, in our case ‘Date’,‘High’ and ‘Low’ attributes are dropped. 8 FEATURE IMPORTANCE
- 10. 10 SO, WHAT ARE THE DIFFERENT MODELS AVAILABLE?
- 11. APPROACHES TO SOLVE ● Linear Regression ● Recurrent Neural Network + LSTM 11
- 12. LINEAR REGRESSION ● Linear regression is an approach for predictive modeling to showcase the relationship between a scalar dependent variable ‘Y’, (in our case, we have ‘Close’ attribute) and one or more independent variables ‘X’ (‘Trading day’ attribute). 12 Close trading days Now, you can predict Closing price along this line
- 13. LINEAR REGRESSION 13 Weights Bias Predicted Close Vector Day Vector Initially, we will take some random weights and bias to initialize. This equation represents this line
- 14. LINEAR REGRESSION 14 ● Initially, we have some random weight and bias, so it will look like this. ● Next step, calculate the error.
- 15. LINEAR REGRESSION 15 ● Initially, we have some random weight and bias, so it will look like this. ● Next step, calculate the error. Mean Squared Error
- 17. LINEAR REGRESSION 17 Mean Squared Error ● we keep on repeating this process till we achieve the minimum error. Weight = Weight - (n x Error ) Bias = Bias - (n x Error )
- 19. RECURRENT NEURAL NETWORK Information related to previous predicted Closing price Today’s opening price Current state hidden layer output Predicted Closing price Current state Hidden layer output
- 20. RECURRENT NEURAL NETWORK So, it does not allow it to learn from past data as was expected. Vanishing Gradient Problem Less effect on the current calculation of weight Day1 Prediction Day2 Prediction Day3 Prediction Day4 Prediction Day5 Prediction Day6 Prediction Day7 Prediction
- 21. RECURRENT NEURAL NETWORK + LSTM LSTM (Long Short Term Memory) LSTMs are explicitly designed to avoid the long-term dependency problem. Remembering information for long periods of time is practically their default behavior, not something they struggle to learn! Recurrent Neural Network LSTM
- 22. RECURRENT NEURAL NETWORK + LSTM LSTM (How it works?) ● The key to LSTM is the Memory cell state which stores the information. It runs straight down the entire chain. ● LSTM has the ability to remove or add information to these cell state, regulated by structures called gates. ● Gates are composed of sigmoid neural net layer and a multiplication operation. ● Sigmoid layer outputs zero or one. Sigmoid layer Multiplication operation
- 23. RECURRENT NEURAL NETWORK + LSTM LSTM (How it works?) ● There are three gates to protect and control the cell states. ○ forget gate ○ input gate ○ output gate forget gate Input gate Output gate
- 24. RECURRENT NEURAL NETWORK + LSTM LSTM (How it works?) ● First, forget gate looks at ht-1 and xt and outputs a number between 0 and 1. ● 1 represents “keep the information” and 0 represents “remove the information”. Current closing Price Past information of this state
- 25. RECURRENT NEURAL NETWORK + LSTM LSTM (How it works?) ● First, forget gate looks at ht-1 and xt and outputs a number between 0 and 1. ● 1 represents “keep the information” and 0 represents “remove the information”. ● Second, input gate decides which values will be updated, in order to do that a tanh layer creates a vector of Ct (bar). ● Combining these two, create an update to the state. required for updating the weight
- 26. RECURRENT NEURAL NETWORK + LSTM LSTM (How it works?) ● First, forget gate looks at ht-1 and xt and outputs a number between 0 and 1. ● 1 represents “keep the information” and 0 represents “remove the information”. ● Second, input gate decides which values will be updated, in order to do that a tanh layer creates a vector of Ct (bar). ● Combining these two, create an update to the state. ● Third, It’s time to update the old cell Ct-1 to Ct. New Cell state
- 27. RECURRENT NEURAL NETWORK + LSTM LSTM (How it works?) ● First, forget gate looks at ht-1 and xt and outputs a number between 0 and 1. ● 1 represents “keep the information” and 0 represents “remove the information”. ● Second, input gate decides which values will be updated, in order to do that a tanh layer creates a vector of Ct (bar). ● Combining these two, create an update to the state. ● Third, It’s time to update the old cell Ct-1 to Ct. ● Fourth, output will based on our cell state. ● a sigmoid layer will decides what parts of the cell state we’re going to output. Output Layer Current hidden layer information
- 28. RECURRENT NEURAL NETWORK + LSTM
- 29. CONCLUSION Linear Regression Recurrent Neural Network + LSTM ● Model does not fit properly ● Model fits properly Stock price prediction is a complex problem and difficult to predict. Machine learning model doesn’t perform well as compared to Deep Learning model.
- 30. 30 THANK YOU!