Saturday, November 23, 2013
Mathematical Modeling
Self Organizing MapOverview and Application in Prediction
Presented By
Decky Aspandi Latif56070701073
Saturday, November 23, 2013
Layout
● Introduction● SOM in Brief
● Basic of SOM● SOM in Modeling and Prediction● Application of SOM in Stock Prices
Prediction● Conclusion
Saturday, November 23, 2013
Introduction● Currently, great need emerges for better
techniques, tools and practices.● Modeling could be applied to various area →
minimize cost & Optimization● Self Organizing Map → ANN(connectionist
paradigms) → support and changes in approaches & modeling technique
● Disparate data analysis in 2 scales, regional and global.
Saturday, November 23, 2013
Self Organizing Map
● Proposed by Tuevo Kohonen (1972)● Unsupervised Neural Network ● Data driven learning process● Reduce dimensions,display similarities
Saturday, November 23, 2013
SOM (Cont..)
● Mapping Nodes to group of class● Selection of Best Matching Unit● Cooperative LearningAlgorithm :
1. Initialize weight of nodes
2. choose random vector
3. examined & select BMU
4. Calculate Neighbourhood
5. Update appropriate weights
6. Repeat step 2 for N times
Y, R
ed, E
leva
tion
,..
X, Blue, Density,..
Y, R
ed, E
leva
tion
,..
X, Blue, Density,..
Saturday, November 23, 2013
SOM → Modeling
● Clustering Capability● Modeling & Prediction
EcologicalModeling
Regional Data Analysis
Prediction
Saturday, November 23, 2013
Application → Prediction
“ Predicting Stock Prices Using a Hybrid Kohonen Self Organizing Map (SOM) “ , Mark & Olatoyosi, 2007
● Main aim → Stock Prices Prediction
● Applied on LucentI Inc, using five years data → 1251 points
● Hybridization of SOM with Multilayer Perceptron
Saturday, November 23, 2013
HSOM → Prediction (cont)
● Hybrid HSOM outperform SOM & BPN● BPN comes inaccurate when price > $60 →
Significant Loss in investment● HSOM has lowest error
(0~12) → Increase in return of Investment (ROI)
Saturday, November 23, 2013
Conclusion
● ANN can be used to enhance and alter the modeling technique
● SOM is an Unsupervised Neural Network● Clustering classes with mapping nodes● Various application of SOM on Modeling &
Simulation → prediction● By collaborating SOM with other method →
greater results.