Transcript Data Mining BS/MS Project

Data Mining BS/MS Project
Decision Trees for
Stock Market Forecasting
Presentation by Mike Calder
Decision Trees
• Used for stock market forecasting
– Classification trees
– Regression trees
• Analysts attempt to predict the value of a
given stock at some point in the future
• The methods can also be used to predict
trends in the stock market as a whole
2
Motivation
• Accurate predictions in the stock market
allow investing companies to thrive
• Identifying attributes that correlate with
success in the stock market may lead to
finding causation
– If causes of success can be controlled, the
economy can be pushed in a good direction
3
Stock Market Challenges
• Training set can be very large
– All stock data over a period of time
• Predicting attributes tend to be binarized
– like we saw when using the ID3 algorithm
• The target attribute (increase/decrease in
stock value) can be numeric or nominal
4
Sample Decision Tree
Taken from (Trader, 2014)
VAR1 – if a stock’s average true range (average max-min) is greater than the
stock’s moving average (average of end-of-day prices)
VAR3 – if a stock’s end-of-day price is greater than the 50-day moving average
VAR4 – if a stock’s end-of-day price divided by the daily range (max-min) is > 0.5
VAR5 – if a stock’s auto-correlation (degree of correlation with respect to its
previous values) return has been > 0 for 5 days
5
Processing Used
•
•
•
•
•
Correlation based feature selection
Numeric attribute binarization
Target regression calculation
N-fold cross-validation (usually n=10)
Decision tree pruning
6
Sample Results
Taken from (Trader, 2014)
The red lines show real data for the 4 stocks,
black lines show the decision tree’s predictions
(y-axis represents the increase/decrease in stock value by percentage)
7
Additional Complexity
• Advanced techniques can be combined
with decision tree construction
– Hierarchical hidden Markov model (HHMM)
has been combined with decision trees for
stock trend prediction
– Other machine learning algorithms have been
used with decision tree methods to forecast
stock market changes as well
8
References
• R. Trader. “Using CART for Stock Market Forecasting”.
Data Science and Trading Strategies. 2014.
• S. Tiwara. “Predicting future trends in stock market by
decision tree rough-set based hybrid system with
HHMM”. International Journal of Electronics and
Computer Science Engineering. 2012.
• T. Zhang. “Stock Market Forecasting Using Machine
Learning Algorithms”. Stanford University. 2012.
9