Recognizing Head and Shoulders Price Pattern in Time Series Data using Self-Organizing Maps

The Hong Kong University of Science and Technology
Department of Computer Science and Engineering

Final Year Thesis Oral Presentation

Title: "Recognizing Head and Shoulders Price Pattern in Time Series Data
using Self-Organizing Maps"

By

Sathish RAGHURAMAN

Abstract

We aim to recognize the Head and Shoulders (H&S) stock pattern in large 
amounts of unlabeled time series data using unsupervised learning. We 
achieve this by performing trend segmentation and clustering using 
Self-Organizing Maps. We use trend segmentation to reduce the 
dimensionality and noise in the original dataset. The input to the SOM are 
features that measure the proportional change in prices in adjoining time 
intervals. As SOMs clusters data based on the similarity of input 
features, we hypothesize that head and shoulders patterns must be grouped 
together. Based on the structure of the head and shoulder pattern, our 
system searches the output neurons to detect those whose weight vectors 
fit the head and shoulder feature template. When this match is found, the 
price movements corresponding to these samples in this cluster are 
predicted to hold the head and shoulders pattern. Lastly, we experiment 
with different cluster sizes to determine the best cluster dimensions for 
H&S pattern recognition.

Date:                   Wednesday, 6 May 2015

Time:                   2:20 - 3:00pm

Venue:                  Room 5503
                        Lifts 25/26

Committee Members:      Prof. James Kwok (Supervisor)
                        Prof. Gary Chan (Reader)