Stock Price Prediction

第8組

李孟潔 111522030

鄭伊涵 111522061

張友安 111526003

張凱名 111522149

Outline

• Introduction

• Method & Architecture

​

• Experimental Result

​

• Additional Experiments

​

• Conclusion

Introduction

Introduction

• Goal:

Use some of DL & ML method to predict open stock price.

​

• Dataset:

training data：Google open stock price from 2012 to 2016.

testing data： 19 days of open stock price in 2017.

​

• Method:

(1) Linear Regression

(2) LSTM

(3) GRU

Method & Architecture

Linear Regression

Property:

• Find the relationship between multiple independent variables and dependent variables.
• According to the number of independent variables, it can be divided into Simple linear regression and multiple regression.

​

Usage:

• Deal with causal sequence like numerical problems or statistical problems.
• Can also be applied to non-linear problems

​

Defect:

• The analysis of causal relationship is not simply found in a linear regression.

LSTM

Property:

• Overcome long-term memory problem in RNN.
• Solve vanishing gradient problem.

​

Usage:

• Good at dealing with long sequence problems prevent from the trace lost in previous information.

​

Defect:

• Very long sequences are still not resolved well.
• training time consuming

GRU

Property:

• The advantage which LSTM has.
• Fewer parameters, reducing the risk of overfitting.

​

Usage:

• Good at dealing with long sequence problems prevent from the trace lost in previous information and improve the efficiency.

​

Defect:

• Very long sequences are still not resolved well.
• No significant difference in training effectiveness compared with LSTM.

Experimental Result

Linear Regression

LSTM

GRU

Compare

Additional Experiments

Logistic Regression

Prophet

Conclusion

Conclusion

• We use both ML and DL method to confirm each other performance apply in the stock price.

​

• Both ML field and DL field method we tried to use had a good performance.

​

• We tried to use numeric sequence which seem to do not have enough significant meaning to predict stock price, and the result prove that the method to predict is work.

​

• Regardless of the ML or DL method, the prediction depends on the past data, so the reliability of the results is still a question.

​

​

​