Trend Forecasting in Financial Time Series Using a Combinational Method of Heuristic Pattern Recognition and Support Vector Machine

EasyChair Preprint 10141

Abstract

Whereas many studies have been done on forecasting different time series, it has always been associated with challenges such as uncertainty. For example, in financial time series, if we want to predict the price, due to the Non-stationary nature of the time series, forecasting will face false regression. To solve this problem, in this research, instead of price forecasting, trend forecasting has been done. In this case, since the subtraction operator has been used to calculate the trend, the effect of Non Stationary nature is removed and the issue of false regression is solved. In this research, using machine learning methods through the return forecasting approach, the trend in financial time series has been predicted. In this research, the effective features of the data of the last 10 years in the foreign stock market for the shares of several different companies have been examined and compared with the Benchmark index of the market, and after creating different machine learning models and maximizing the accuracy of the results, a satisfying application has been extracted to be used as an effective trading tool for traders. To train the model, random forests algorithm and Support Vector Machine, Feature Selection and Heuristic algorithms have been used. The evaluation of this model on the Foreign Stock market in recent years shows that not only the presented system can make effective predictions, but it is largely robust to market fluctuations and performs better than other existing methods.

Keyphrases: Forecasting, financial time series, machine learning, trend prediction

@booklet{EasyChair:10141,
  author    = {Fatemeh Khazaeni and Mohammad Amin Shayegan},
  title     = {Trend Forecasting in Financial Time Series Using a Combinational Method of Heuristic Pattern Recognition and Support Vector Machine},
  howpublished = {EasyChair Preprint 10141},
  year      = {EasyChair, 2023}}