Combined ensemble multi-class SVM and fuzzy NSGA-II for trend forecasting and trading in Forex markets

Abstract

Foreign exchange (Forex) market is the biggest currency exchange market in the world. Existing trading systems in Forex markets based on technical analysis use crisp technical indicators to provide Buy/Sell signals to the trader, only when the indicator value crosses a given threshold level. This strict and noise-sensitive condition can be replaced through uncertainty handling of indicators using fuzzy numbers to generate Buy/Sell signals with fuzzy memberships functions. To achieve this purpose, this paper presents a combined technique based on ensemble multi-class support vector machine (EmcSVM) and fuzzy NSGA-II for efficient trend classification and trading in Forex markets. At first, EmcSVM is used to forecast and classify the future market trend into uptrend, sideway, and downtrend. Then, NSGA-II is applied to optimize the hyperparameters of the proposed fuzzy trading system comprising multiple AND-OR Buy/Sell technical rules for uptrend/downtrend markets. The hyperparameters include indicator selection within each rule, importance weights of the different rules, and final decision thresholds for Buy/Sell models, while the objective is to maximize average return on investment (ROI) and minimize average draw-down of all transactions. The proposed method has been successfully developed and tested on real data from the Forex market for EUR/USD currency pair in a 6-year timeframe from 2014 to 2019. Obtained results show that the proposed method outperforms the existing crisp trading systems, with 80.8% precision, 72.4% recall, 94.1% annual ROI, and 0.58% draw down.