From: Forex market forecasting using machine learning: Systematic Literature Review and meta-analysis
Studies | Machine learning model (Algorithm) | Dataset | Validation technique | Evaluation metrics | Formula | Baseline models | |
---|---|---|---|---|---|---|---|
Currency pair | Period | Â | |||||
[P1] | Forex Loss Function (FLF) With Long Short-Term Memory | EURUSD | The H4 candles EURUSD data was collected from June 2015 to Sep 2018 using an MT4 tool | Percentage split (60–40) | Mean Absolute Error (MAE of ARIMA), MAE of FLF-LSTM, MAE of FB Prophet | \(MAE=\frac{1}{n}\sum_{i=1}^{n}({Y}_{i}-\widehat{Y})\) | Facebook Prophet |
[P2] | Knowledge guided artificial neural network (KGANN) | conversion rates from 1US$ to Pound and Rupees | NULL | Percentage split | MAPE | \(\mathrm{MAPE}(\mathrm{k})=\frac{\sum_{m-1}^{M}PE(m)}{M}\) | LMS FLANN |
[P3] | Long Short-Term Memory Networks | AUDUSD | January 1, 2016, to March 31, 2019 | Percentage split | root mean square error (RMSE) and the mean absolute percentage error (MAPE), | \(RMSE=[\frac{1}{n}{\sum }_{i=1}^{n}({y}_{i}-{\widehat{y}}_{i}{)}^{2}{]}^\frac{1}{2}\) \(MAPE=\frac{1}{n}{\sum }_{i=1}^{n}\left|\frac{{y}_{i}-\widehat{{y}_{i}}}{{y}_{i}}\right|\) | NOT AVAILABLE |
[P4] | Deep Neural Network (LSTM, CNN) | (EUR/USD), (GBP/USD), (USD/JPY), AZN/USD, AMD/USD, BYR/USD, MDL/USD, UAH/USD, GEL/USD | 2000 to 2015(For each data set we train models for daily, monthly, and quarterly predictions.) | NA | NA | NA | ARIMA, ETS, NN (Single Layer ( |
[P5] | Chaos, Neural Network and Particle Swarm Optimization (Hybrid) | USD-JPY, USDGBP and USDEUR | The daily data of USD-JPY, USDGBP (1–1-1993 to 31–12-2013) and USD-EUR (3–1-200 to 31–12-2013 | Percentage split (80–10) | Mean Squared Error (MSE) and Mean Absolute Percentage Error | \(MSE=\frac{\sum_{t=1}^{k}(y\left(t\right)-\dot{y(t){)}^{2}}}{k}\) \(MAPE\frac{100}{k}{\sum }_{t=1}^{k}\left|\frac{y\left(t\right)-\dot{y(t)}}{y(t)}\right|\) | MLP, GRNN, GMDH and PSO |
[P6] | Neural networks (NN) LSTM | EUR/USD | daily exchange rate data from 1.4.1971 until 9.5.2019 | NA | Mean Square Error (MSE) and mean absolute error (MAE) | \(MSE=\frac{\sum_{t=1}^{k}(y\left(t\right)-\dot{y(t){)}^{2}}}{k}\) \(MAE={\sum }_{t=1}^{k}\left|\frac{y\left(t\right)-\dot{y(t)}}{y(t)}\right|\) | NULL |
[P7] | Support Vector Regressor (SVR), Artificial Neural Network (ANN), Long Short-Term Memory (LSTM), Neural Network with Hidden Layers | USD/EUR, USD/JPY, USD/GBP, USD/AUD, USD/CAD, USD/CHF, USD/CNY, USD/SEK, USD/NZD, USD/MXN and USD/INR | The data used to predict are daily data of the currency rate of the 1980s to December 2018 period | Percentage split (80–20) | Loss function: Mean Square Error, Optimizer: Adam The coefficient of determination ( | \(MSE=\frac{\sum_{t=1}^{k}(y\left(t\right)-\dot{y\left(t\right){)}^{2}}}{k}\) \({R}^{2}=1-\frac{\sum_{i}({y}_{i}-{f}_{i} {)}^{2}}{\sum_{i}({y}_{i}-{\overline{y} }{)}^{2}}\) \({y}_{i}:n values of the dataset\) \({f}_{i}\): Predicted values \(\overline{y }:means of observed data\) | NULL |
[P8] | GRU-LSTM | EUR/USD, GBP/USD, USD/CAD, and USD/CHF | January 1, 2017 to December 31, 2018(10-min timeframe, and 30 min timeframe using the data from January 1, 2019 to June 30, 2020 | Percentage split (80–20) | MSE, RMSE, MAE and R-square score, | \(MSE=\frac{\sum_{t=1}^{k}(y\left(t\right)-\dot{y(t){)}^{2}}}{k}\) \(MAE{\sum }_{t=1}^{k}\left|\frac{y\left(t\right)-\dot{y(t)}}{y(t)}\right|\) \({R}^{2}=1-\frac{\sum_{i}({y}_{i}-{f}_{i} {)}^{2}}{\sum_{i}({y}_{i}-{\overline{y} }{)}^{2}}\) | Standalone GRU model, a standalone LSTM model, and simple moving average (SMA) |
[P9] | Reinforcement Learning (Neural networks with three hidden layers of ReLU neurons are trained as RL agents under the Q-learning algorithm) | EUR/USD | 2010 to 2017 | Percentage split (Rolling window) | NA | NA | NA |
[P10] | Neural networks (MLP, RNN, GRU, and LSTM) | EUR/USD, USD/JPY, USD/CHF, GBP/USD, USD/CAD, and AUD/USD | GBP/USD (2000 to 2019) | K-Fold cross-validation 80–20 | MAE | \(MAE={\sum }_{t=1}^{k}\left|\frac{y\left(t\right)-\dot{y(t)}}{y(t)}\right|\) RMSE | Random Forest, AdaBoost, XGBoost and Support vector machine |
[P11] | Logistic/Linear Regression, Regularized Logistic/Linear Regression, Support Vector Machines/Regression (SVM/SVR), Gradient Boosting Classifier/Regression (GBC/GBR), Neural Networks (NN) | USDMXN | The dataset spans between the first week of January 2003 and the second week of November 2018 | Percentage split (60–40) | NA | NA | NA |
[P12] | Linear Regression Line, Artificial Neural Network and Dynamic Time Warping Algorithms | AUDUSD and EURUSD | from the beginning of 2012 until the end of 2012 | Percentage split (70–30) | Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) | \(MAE={\sum }_{t=1}^{k}\left|\frac{y\left(t\right)-\dot{y(t)}}{y(t)}\right|\) \(RMSE=\sqrt{\frac{\sum_{i=1}^{N}{\Vert y\left(i\right)-\widehat{y}(i)\Vert }^{2}}{N}}\) | NA |
[P13] | Neural networks | USD/EUR | July 1, 2010 to March 23, 2011 | NA | NA | NA | NA |
[P14] | Empirical Mode Decomposition (EMD) and Long Short-Term Memory (LSTM) | USD/CNY, EUR/CNY, and USD/EUR | January 1, 2000 to April 17, 2018 | Percentage split (94–6) | Mean square error (MSE), mean absolute error (MAE), and mean absolute percentage error (MAPE), | \(MSE=\frac{\sum_{t=1}^{k}(y\left(t\right)-\dot{y(t){)}^{2}}}{k}\) \(MAE={\sum }_{t=1}^{k}\left|\frac{y\left(t\right)-\dot{y(t)}}{y(t)}\right|\) \(MAPE=\frac{100}{k}{\sum }_{t=1}^{k}\left|\frac{y\left(t\right)-\dot{y(t)}}{y(t)}\right|\) | ARIMA, ANN, LSTM |
[P15] | LSTM, RNN | EUR/USD | June 1993 to March 2018 | Percentage split (80–20) | RMSE and MAE | \(RMSE=\sqrt{\frac{\sum_{i=1}^{N}{\Vert y\left(i\right)-\widehat{y}(i)\Vert }^{2}}{N}}\) \(MAE={\sum }_{t=1}^{k}\left|\frac{y\left(t\right)-\dot{y(t)}}{y(t)}\right|\) | NA |
[P16] | Hybrid (RNN AND CNN) | EURUSD, AUDUSD, XAUUSD, GBPJPY,EURJPY,GBPUSD,USDCHF,USDJPY,USDCAD | June 2008 to May 2018 | Percentage split | NA | NA | LSTM or CNN deep neural network |
[P17] | Recurrent Neural Networks | CHF EUR GBP JPY | 22.07.1998 to 02.09.2001 Daily timeframe | NA | R squared Mean Squared Error | \({R}^{2}=1-\frac{\sum_{I}({y}_{i}-{\widehat{y}}_{i}{)}^{2}}{\sum_{I}({y}_{i}-{\overline{y} }_{i}{)}^{2}}\) | NA |
[P18] | NEURAL NETWORK | USD / EUR | 23.04.2012—04.05.2012 | Percentage split (95–5) | NN1 NN2 NN3 | NA | NA |
[P19] | LSTM | GBP/USD, EUR/GBP, AUD/USD and CAD/CHF | 15Â min interval data from Jan 2005 to Sep 2017 for training and data from Oct 2017 to Sep 2020 for testing | Percentage split | Square error (MSE), root mean squared error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE) | \(MSE=\frac{\sum_{t=1}^{k}(y\left(t\right)-\dot{y\left(t\right){)}^{2}}}{k}\) \(RMSE=\sqrt{\frac{\sum_{i=1}^{N}{\Vert y\left(i\right)-\widehat{y}\left(i\right)\Vert }^{2}}{N}}\) \(MAE={\sum }_{t=1}^{k}\left|\frac{y\left(t\right)-\dot{y\left(t\right)}}{y\left(t\right)}\right|\) \(MAPE=\frac{100}{k}{\sum }_{t=1}^{k}\left|\frac{y\left(t\right)-\dot{y(t)}}{y(t)}\right|\) | Simple recurrent neural network (RNN) |
[P20] | Recurrent neural Networks (FNN SRNN LSTM GRU) | EUR/USD GBP/USD USD/JPY USD/CHF | January 4, 1971 until August 25, 2017 | We divide each time series into study periods, scale the training data, and create input sequences and target variable values. Percentage split | Forecast Accuracy: logarithmic loss Predictive Accuracy (AUC) | NA | NA |
[P21] | Long short-term memory (LSTM) and autoencoder models | currency-related volatility indices, namely, the BPVIX, JYVIX, and EUVIX | January of 2010 to December of 2019 | K-fold cross-validation | MSE RMSE MAE MPE MAPE | NA | NA |
[P22] | REGRESSION TECHNIQUES | INR/USD and INR/EUR | November 9th, 2016 to July 31st, 2017 | Data is divided into two parts: training and testing (Percentage split). K-fold cross-validation technique is used for dynamic data partitioning to make the prediction more accurate | MAPE | \(MAPE=\frac{\sum_{i=0}^{n}\left|{Y}_{a,i}-{Y}_{p,i}\right|}{n}\) | NA |
[P23] | Long short-term memory (LSTM | EUR/USD | January 2013–January 2018 | Percentage split (80–20) | accuracy | \(\mathrm{ProfitAccuracy}=\frac{sum of true}{ sum of false}\) | NA |
[P24] | Support Vector Machine (SVM) | EUR/USD | 01/01/2013 to 30/09/2016 | cross-validation method | accuracy rates | NA | No_SVM |
[P25] | Deep LSTM with Reinforcement Learning Layer | EUR/USD, GBP/USD, EUR/GBP | 2004–2018 | Percentage split (70–30) | accuracy | \(ROI=\frac{100*\left(Gain-Investment)\right)}{\begin{array}{c}Investment\\ \end{array}}\) \(MD=\frac{\left|Balance Valley-Balance Peak\right|}{BalancePeak}\) | Grid Trading System Trading System Threshold Strategy |
[P26] | Bagging Trees, SVM, MLP, RBF | USD/YEN, USD/EGP, EURO/EGP and EURO/SAR | April 2003 to August 2010 excluding the weekends | Percentage split | NA | NA | NA |
[P27] | artificial neural network | AUD/USD, CAD/USD, CHF/USD, EUR/USD, GBP/USD | January 2001 to November 2013 | NA | MSE, the MAD, and the MAPE measures | \(MSE=\frac{1}{n}{\sum }_{i=1}^{n}{\left({RV}_{i}-{\widehat{{\sigma }_{i}}}^{2}\right)}^{2}\) \(MAPE=\frac{1}{n}{\sum }_{i=1}^{n}\left(\frac{{RV}_{i}-{\widehat{{\sigma }_{i}}}^{2}}{{RV}_{i}}\right)\) \(MAD=\frac{1}{n}{\sum }_{t=1}^{n}\left|{RV}_{i}-{\widehat{{\sigma }_{i}}}^{2}\right|\) | Autoregressive conditional heteroskedasticity (GARCH), GJR, and Asymmetric Power Generalized Autoregressive Conditional Heteroskedasticity (APGARCH) models |
[P28] | Elastic Network Model (ENMX) | AUDCAD, AUDJPY , AUDNZD, AUDUSD, CADJPY EURAUD EURCAD EURGBP EURJPY EURNZD EURUSD GBPAUD GBPCAD GBPJPY GBPNZD GBPUSD NZDCAD NZDJPY NZDUSD USDCAD USDJPY | 01.01.2015 (23:00Â h) to 07.28.2017 (18:00Â h), | NA | RMSE PF | \({RMSE}_{c}=\sqrt{\frac{1}{n}\sum_{i=1}^{n}\left\{{\widehat{y}}_{i,t+H}-{y}_{i,t+H{\}}^{2}}\right.}\) \(PF=\frac{gross profit}{gross losses}\) | Simple random walk (RW) and vector autoregressive (VAR) model |
[P29] | Extreme Learning Machines (ELMs) and the Jaya optimization technique | USD to INR USD to EURO | 04/05/2000–03/06/2016 14/09/2001–03/06/2016 | Percentage split | MSE MAPE MAE Theil’s U ARV | \(MSE=\frac{1}{n}{\sum }_{i=1}^{n}({A}_{i}-{P}_{i}{)}^{2}\) \(MAPE=\frac{\sum_{i=1}^{N}\left|\frac{{A}_{i}-{P}_{i}}{N}\right|}{N} \times 100\) \(MAE=\frac{1}{N}\sum_{i=1}^{N}\left|{A}_{i}-{\overline{P} }_{i}\right|\) \(Thei{l}{^{\prime}}s\,U=\frac{\sqrt{\frac{1}{n}{\sum }_{i=1}^{n}({A}_{i}-{P}_{i}{)}^{2}}}{\sqrt{\frac{1}{n}{\sum }_{i=1}^{n}({{A}_{i})}^{2}} + \sqrt{\frac{1}{n}{\sum }_{i=1}^{n}({{P}_{i})}^{2}}}\) \(ARV=\frac{{\sum }_{i=1}^{n}({P}_{i}-{A}_{i}{)}^{2}}{{\sum }_{i=1}^{n}({P}_{i}-{\overline{A} }_{i}{)}}\) | NA |
[P30] | Recurrent Cartesian Genetic Programming evolved Artificial Neural Network (RCGPANN | Data of US currency obtained from Australian Reserved Bank Japanese Yen, New Zealand Dollars, Canadian Dollars, Korean Won, and Indonesian Rupiah | 500Â days 1000Â days from 1st February 2003 | Percentage split | MAPE | \(MAPE=\frac{1}{N}{\sum }_{i=1}^{n}\left(\frac{\left|{L}_{F}-{L}_{A}\right|}{{{L}_{A}}}\right)X 100\) | NA |
[P31] | Bayesian Compressed Vector Autoregression | AUD-JPY, CAD-CHF, CAD-JPY, EURDKK, EUR-MXN, and EUR-TRY | 7th February 2018 through 2 august 2018 | Percentage split (70–30) | BCVAR-Mean square forecasting error (MSFE), Mean absolute forecasting error (MAFE) | NA | NA |
[P32] | Modified fuzzy relational model (MFRM); | USDCHF | Jan.2012 to Dec. 2013 Daily time frame | Percentage split (3.3–96.7) | MSE, MAPE, and RMSE | \(MSE=\frac{\sum_{t=1}^{k}(y\left(t\right)-\dot{y\left(t\right){)}^{2}}}{k}\) \(MAPE=\frac{100}{k}{\sum }_{t=1}^{k}\left|\frac{y\left(t\right)-\dot{y(t)}}{y(t)}\right|\) \(RMSE=\sqrt{\frac{\sum_{i=1}^{N}{\Vert y\left(i\right)-\widehat{y}\left(i\right)\Vert }^{2}}{N}}\) | Multi-layer perceptron’s, radial basis function neural networks and artificial fuzzy interface systems |
[P33] | Multilayer Perceptron (MLP) and Adaptive Neuro- Fuzzy Inference System (ANFIS) | USDINR | 23rd October 1993 to 23rd October 2014 (Daily period) | NA | Root Mean Square Error (RMSE) and D-stat (success of prediction in terms of direction | NA | NA |
[P34] | Adaptive Network-based Fuzzy Inference 19 System (ANFIS) | EUR/USD GBP/USD USD/JPY USD/CHF | January 1, 2011, to February 28, 2014 | Percentage split | Root Mean Squared Error (RMSE), the Standard Error of the Mean (SEM), the Mean Absolute Percentage Error (MAPE) and the Mean Absolute Error (MAE) | \(RMSE=\sqrt{\frac{1}{N}\sum_{i=1}^{N}{({Actual}_{i}-{Forecast}_{i})}^{2}}\) \(SEM=\frac{\sigma }{\sqrt{N}}=\sqrt{\frac{1}{N(N-1)}\sum_{i=1}^{N}{({X}_{i}-{\widehat{X}}_{i})}^{2}}\) \(MAPE=\left(\frac{1}{N}\sum_{i=1}^{n}\left|\frac{{Actual}_{i}-{Forecast}_{i}}{{Actual}_{i}}\right|\right)*100\) \(MAE=\frac{1}{N}\sum_{i=1}^{N}\left|{Actual}_{i}-{Forecast}_{i}\right|\) | NA |
[P35] | Synthesis of fuzzy logic and the Dempster–Shafer theory of evidence (SYSTEM) | EUR/USD, GBP/USD EUR/CHF USD/CHF | The quotations of currencypairsEUR/USD (timeframes1Hand4H), GBP/USD (30Mand4H), EUR/CHF(1H and4H) orazUSD/CHF (15Mand1H) from01.11.2013to31.12.2014wereused | NA | NA | NA | NA |
[P36] | Support Vector Machine with Genetic Algorithms | EUR/USD | 1st of January of 2003 and 1st of January of 2015 | Percentage split | The Backtesting approach is based on a Sliding Window | NA | Random Walk, B&H, S&H, and a Static GA approach |
[P37] | Support Vector Regression (SVR)-Wavelet Adaptive Model | AUD-JPY, CHF-JPY, EUR-JPY, GBP-JPY and EUR-CHF | 01/01/2003 to 12/30/2014, around 3000 values per series with intervals of 1Â day, 1Â h and 15Â min | NA | Mean Error (ME), Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Percentage Error (MPE), Mean Absolute Percentage Error (MAPE), Mean Absolute Scaled Error (MASE), Autocorrelation of errors at lag 1 (ACF1), Akaike Information Criterion (AIC), Second-Order AIC (AICc) and Bayesian Information Criterion (BIC) | NA | ARIMA and ARFIMA Model |
[P38] | Deep recurrent neural network (DRNN)-ARIMA | EUR/USD | data of EUR/USD between 2013 and 2015 were used to train the model and price data of EUR/USD between 2016 and 2017 were used to test and evaluate the model | Percentage split | MAPE, MSE | \(MAPE=\left(\frac{100\%}{N}\sum_{i=1}^{n}\left|\frac{{A}_{i}-{F}_{i}}{{A}_{i}}\right|\right)\) \(MSE=\frac{1}{N}\sum_{i=1}^{N}{({A}_{i}-{F}_{i})}^{2}\) | NA |
[P39] | NEURAL NETWORK | USD/EUR | 23.04.2012–04.05.2012 | Percentage split | NN1 NN2 NN3 | NA | NA |
[P40] | LSTM | EUR / USD | 2014 to March 2020 | Percentage split (80–20) | RMSE | \(RMSE=\sqrt{\frac{1}{N}\sum_{i=1}^{N}{({Actual}_{i}-{Predict}_{i})}^{2}}\) | ARIMA |
[P41] | KNN with Dynamic Time Warping AS Distance Function | USD/JPY | Dec 4th of 1971 to Dec 24th of 2012 | Percentage split (70–30) | RMSE | \(RMSE = \sqrt[2]{{\frac{1}{N}\sum\limits_{{i = 1}}^{N} ( x_{i} - \mathop {x_{i} }\limits^{'} )^{2} }}\) | Artificial Neurl Network Multilayer Perceptron |
[P42] | Chaos-based support vector regressions | (EUR/USD), (GBP/USD), (NZD/USD), (AUD/USD), (JPY/USD) and (RUB/USD) | January 3, 2005 to December 31, 2007, (daily timeframe) | Percentage split (80–20) | MSE, RMSE and MAE | \(MSE=\frac{1}{N}\sum_{i=1}^{N}{({r}_{i}-{t}_{i})}^{2}\) \(RMSE=\sqrt{\frac{1}{N}\sum_{i=1}^{N}{({r}_{i}-{t}_{i})}^{2}}\) \(MAE=\frac{1}{N}\sum_{i=1}^{N}\left|{r}_{i}-{t}_{i}\right|\) | SVR and BPNN |
[P43] | Recurrent Neural Network (RNN) of the Elman type, and Long Short-Term Memory (LSTM) | EUR/USD | 2 January 1998, to 31 December 2019 Daily prices | Percentage split | Mean Square Error (MSE), Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Median Absolute Error (MdAE), Mean Absolute Percentage Error (MAPE), Symmetric Mean Absolute Percentage Error (sMAPE), Root Mean Square Percentage Error (RMSPE), Root Median Square Percentage Error (RMdSPE), Median Absolute Percentage Error (MdAPE), and Symmetric (MdAE), Mean Absolute Percentage Error (MAPE), | NA | NA |
[P44] | Decision Trees (CART, C4.5) | EURUSD GBPUSD USDJPY | UNKNOWN | Random-sub sample | Confusion Metrix | NULL | CART, C4.5 |
[P45] | Support Vector Machine (SVM) and a hybrid form of Genetic Algorithm-Neural Network (GA-NN) | Â | 2007 to September 2012 Hourly data | Percentage split | RSME | \(RMSE=\sqrt{\frac{1}{N}\sum_{i=h}^{N}{({y}_{i}-{\widehat{y}}_{i})}^{2}}\) | ASTAR model, SVM, and GANN |
[P46] | Ensemble of Neural Networks based on Conventional Technical Indicators | EUR / USD | October 2018 | NA | NA | NA | NA |
[P47] | Naive Bayes Classifi er | EUR/USD | January 1st, 2013 till March 9th, 2017(hourly rate) | Percentage split and k-fold CV (80–20) | Accuracy, precision and recall | NA | NA |
[P48] | Support Vector Machines | EUR/USD | 01/01/2013 to 30/09/2016 Time frames of M1, M5, M15, H1, D1 | Percentage split | MSE | \(MSE=\frac{1}{N}\sum_{i=1}^{N}{({r}_{i}-{t}_{i})}^{2}\) | NA |
[P49] | Chaos Theory (Random Forest, Classification and regression tree) and Multivariate Adaptive Regression Splines | JPY/USD, GBP/USD, and EUR/USD | JPY/USD, GBP/USD are from January 1, 1993 to December 31, 2013 and EUR/USD from January 3, 2000 to December 31, 2013 | Percentage split | MSE, MAPE | \(MSE=\frac{\sum_{t=1}^{N}{E}_{t}^{2}}{N}\) \(MAPE=\frac{100}{N}\sum_{t=1}^{N}\left|\frac{{E}_{t}}{{y}_{t}}\right|\) | NA |
[P50] | Support Vector Machine (SVM) with the Radial Basis Function (RBF) kernel and K-Means clustering algorithms | EUR vs USD, GBP vs JPY, USD vs JPY, GBP vs USD and EUR vs JPY | NA | Percentage split | NA | NA | NA |
[P51] | Nonlinear Autoregressive models with eXogenous input (NARX) Neural Network with Bagging | GBPUSD and EURUSD | 1Hour time frame | Percentage split | RMSE | \(RMSE=\sqrt{\frac{1}{N}\sum_{i=1}^{N}{\left({y}_{i}-{t}_{i}\right)}^{2}}\) | NA |
[P52] | Geometry Sensitive Neural Networks | EURUSD | 2009–11-5–22:15 to 2009–11-20–10:15, using a 15-min time frame | Percentage split | NA | NA | NA |
[P53] | linear regression line, artificial neural network (ANN), dynamic time warping (DTW) | AUDUSD and EURUSD | from 2012 and 2013 | Percentage split | absolute error (MAE) and root mean square error (RMSE) | \(MAE=\frac{1}{N}\sum_{i=1}^{N}\left|{r}_{i}-{t}_{i}\right|\) \(RMSE=\sqrt{\frac{1}{N}\sum_{i=1}^{N}{\left({y}_{i}-{t}_{i}\right)}^{2}}\) | SVM, ANN, Statistic method |
[P54] | Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) and Improved Firefly Algorithm-Long Short-Term Memory (IFALSTM) | AUD/USD, EUR/USD, GBP/USD | from January 1, 2010 to December 30, 2019 | Percentage split | Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE) to | \(RMSE=\sqrt{\frac{1}{N}\sum_{i=1}^{N}{\left({Y}_{i}-{\widehat{Y}}_{i}\right)}^{2}}\) \(MAE=\frac{1}{N}\sum_{i=1}^{N}\left|{Y}_{i}-{\widehat{Y}}_{i}\right|\) \(MAPE=\frac{1}{N}\sum_{i=1}^{N}\frac{{Y}_{i}-{\widehat{Y}}_{i}}{{Y}_{i}}\) | LSTM, CEEMDAN-LSTM, |
[P55] | Multichannel-LSTM | EURUSD and EURAUD | 30Â min timeframe | Percentage split | Accuracy, F1, Precision, and Recall | NA | GRU, Simple RNN, and Single-Channel LSTM, Random Forest and SVM |
[P56] | RNN, GRU, LSTM and MLP | EUR/USD, USD/JPY, USD/CHF, GBP/USD, USD/CAD and AUD/USD | from August 28th 2004 to May 18th 2020) | Percentage split | RMSE MAE | NA | Random forest, AdaBoost, XGBoost and SVM |
[P57] | ANN, ANNMA, SVM, and CNN | EUR/USD, GBP/USD, and JPY/USD | 2010 to 2015 Daily time frame | Percentage split and Ten-fold cross-validation | Accuracy | NA | (ARIMA and ETS) MC (Majority Class) |
[P58] | Wavelet Denoised-ResNet CNN and LightGBM | USDJPY | 2019–01-01 to 2020–06-10 5-min time frame | Percentage split (80–20) | MAE, RMSE | \(MAE=\frac{1}{M}\sum_{i=1}^{M}\left|{x}_{i}-{y}_{i}\right|\) \(RMSE=\sqrt{\frac{1}{M}\sum_{i=1}^{M}{\left({x}_{i}-{\widehat{y}}_{i}\right)}^{2}}\) | NA |
[P59] | ARIMA, Neural Network and Fuzzy Neuron | USD Vs INR GBP Vs INR EURO Vs INR YEN Vs INR | Daily RBI reference exchange rates from January 2010-April 2015 | NA | NA | NA | ARIMA, Neural Network and Fuzzy Neuron |
[P60 | SVM and fuzzy NSGA-II | EUR/USD | 6-year timeframe from 2014 to 2019 Daily timeframe | Percentage split | Accuracy Precision Recall | \(\mathrm{Accuracy }=\frac{No. Correctly Classified Samples }{No. Test Samples}\) \({Precision}_{c} =\frac{{TruePositive}_{C} }{{TruePositive}_{C} + {FalsePositive}_{C}}{\mathrm{Recall}}_{c} =\frac{{TruePositive}_{C} }{{TruePositive}_{C} +{FalsePositive}_{C}}\) | SVM-GA |