PREDICTION OF TIME SERIES DATA USING ANT SYSTEM ALGORITHM (Case Study of Stock Price Prediction)

• Main Author: SEPTEM RIZA (NIM 23205301), LALA
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/8450
• Institution: Institut Teknologi Bandung

Prediction of time series data is often conducted in finance, biology, astronomy, medical, meteorology, etc. The aim of this prediction is to obtain early clue about the future condition so that precise response could be taken. This research would examine one of time series data prediction problem in finance, i.e. prediction of stock price fluctuation. Two approximations commonly conducted by asset analysis expert in predicting the stock price are technical analysis based on histories data and fundamental analysis based on macroeconomic and the company condition. This thesis develops a model clustering those two approximations. This research uses similar sequence matching (SSM) method to investigate the pattern of histories data. SSM method conducts pattern investigation and classification in previous data based on pattern sample that has been determined, whereas Euclidean distance is used as parameter to measure similarity. Furthermore, this research also uses max – min ant system method to combine SSM method with fundamental factors i.e. company condition, macroeconomic and non economic factor involved. There are four significant aspects in building ant system method, i.e. conducting graph construction representing the faced problem, developing heuristic function model and transition rule, developing pheromone updating model, and method used as discharge iteration criteria. Heuristic function implemented in ant system method is a representation of influence of stock price fluctuation data in past and influence of present condition i.e. measurement of company condition factor (price earning ratio, dividend yield, etc), macroeconomic condition (inflation level, interest level, oil price, etc), and non-economy condition (domestic condition, fluctuation of foreign stock index, etc). The outputs of this research are model and software to predict stock price. The simulation which has been carried out shows satisfying result to predict pattern of the stock price fluctuation and give tolerate error to predict return value/stock price.