Improving Deep Learning for Forecasting Accuracy in Financial Data

Financial forecasting is based on the use of past and present financial information to make the best prediction of the future financial situation, to avoid high-risk situations, and to increase benefits. Such forecasts are of interest to anyone who wants to know the state of possible finances in the future, including investors and decision-makers. However, the complex nature of financial data makes it difficult to get accurate forecasts. Artificial intelligence, which has been shown to be suitable for analyzing very complex problems, can be applied to financial forecasting. Financial data is both nonlinear and nonstationary, with broadband frequency features. In other words, there is a large range of fluctuation, meaning that predictions made only using long short-term memory (LSTM) are not enough to ensure accuracy. This study uses an LSTM model for analysis of financial data, followed by a comparison of the analytical results with the actual data to see which has a larger root-mean-square-error (RMSE). The proposed method combines deep learning with empirical mode decomposition (EMD) to understand and predict financial trends from financial data. The financial data for this study are from the Taiwan corporate social responsibility (CSR) index. First, the EMD method is used to transform the CSR index data into a limited number of intrinsic mode functions (IMF). The bandwidth of these IMFs becomes narrower, with regular cyclic, periodic, or seasonal components in the time domain. In other words, the range of fluctuation is small. LSTM is a good way to forecast cyclic or seasonal data. The forecast result is obtained by adding all the IMFs together. It has been verified in past studies that only the LSTM and LSTM combined with the EMD can be used. The analytical results show that smaller RMSEs can be obtained using the LSTM combined with EMD compared to real data.

Download Full-text

DeepSimulator: a deep simulator for Nanopore sequencing

10.1101/238683 ◽

2017 ◽

Cited By ~ 1

Author(s):

Yu Li ◽

Renmin Han ◽

Chongwei Bi ◽

Mo Li ◽

Sheng Wang ◽

...

Keyword(s):

Deep Learning ◽

De Novo ◽

Real Data ◽

Electrical Current ◽

Complex Nature ◽

Main Task ◽

Nanopore Sequencing ◽

Base Calling ◽

Context Dependent ◽

Sequencing Procedure

ABSTRACTMotivationOxford Nanopore sequencing is a rapidly developed sequencing technology in recent years. To keep pace with the explosion of the downstream data analytical tools, a versatile Nanopore sequencing simulator is needed to complement the experimental data as well as to benchmark those newly developed tools. However, all the currently available simulators are based on simple statistics of the produced reads, which have difficulty in capturing the complex nature of the Nanopore sequencing procedure, the main task of which is the generation of raw electrical current signals.ResultsHere we propose a deep learning based simulator, DeepSimulator, to mimic the entire pipeline of Nanopore sequencing. Starting from a given reference genome or assembled contigs, we simulate the electrical current signals by a context-dependent deep learning model, followed by a base-calling procedure to yield simulated reads. This workflow mimics the sequencing procedure more naturally. The thorough experiments performed across four species show that the signals generated by our context-dependent model are more similar to the experimentally obtained signals than the ones generated by the official context-independent pore model. In terms of the simulated reads, we provide a parameter interface to users so that they can obtain the reads with different accuracies ranging from 83% to 97%. The reads generated by the default parameter have almost the same properties as the real data. Two case studies demonstrate the application of DeepSimulator to benefit the development of tools in de novo assembly and in low coverage SNP detection.AvailabilityThe software can be accessed freely at: https://github.com/lykaust15/deep_simulator.

Download Full-text

Deep Learning Based on Multi-Decomposition for Short-Term Load Forecasting

Energies ◽

10.3390/en11123433 ◽

2018 ◽

Vol 11 (12) ◽

pp. 3433 ◽

Cited By ~ 6

Author(s):

Seon Kim ◽

Gyul Lee ◽

Gu-Young Kwon ◽

Do-In Kim ◽

Yong-June Shin

Keyword(s):

Deep Learning ◽

Short Term Memory ◽

Load Forecasting ◽

Small Scale ◽

Intrinsic Mode Functions ◽

Short Term ◽

Term Memory ◽

Load Profile ◽

Memory Network ◽

Long Short Term Memory

Load forecasting is a key issue for efficient real-time energy management in smart grids. To control the load using demand side management accurately, load forecasting should be predicted in the short term. With the advent of advanced measuring infrastructure, it is possible to measure energy consumption at sampling rates up to every 5 min and analyze the load profile of small-scale energy groups, such as individual buildings. This paper presents applications of deep learning using feature decomposition for improving the accuracy of load forecasting. The load profile is decomposed into a weekly load profile and then decomposed into intrinsic mode functions by variational mode decomposition to capture periodic features. Then, a long short-term memory network model is trained by three-dimensional input data with three-step regularization. Finally, the prediction results of all intrinsic mode functions are combined with advanced measuring infrastructure measured in the previous steps to determine an aggregated output for load forecasting. The results are validated by applications to real-world data from smart buildings, and the performance of the proposed approach is assessed by comparing the predicted results with those of conventional methods, nonlinear autoregressive networks with exogenous inputs, and long short-term memory network-based feature decomposition.

Download Full-text

Development of a Model for Predicting the Direction of Daily Price Changes in the Forex Market Using Long Short-Term Memory

International Journal of Machine Learning and Computing ◽

10.18178/ijmlc.2021.11.1.1015 ◽

2021 ◽

Vol 11 (1) ◽

pp. 61-67

Author(s):

Watthana Pongsena ◽

◽

Prakaidoy Sitsayabut ◽

Nittaya Kerdprasop ◽

Kittisak Kerdprasop ◽

...

Keyword(s):

Machine Learning ◽

Time Series ◽

Deep Learning ◽

Short Term Memory ◽

Series Data ◽

Short Term ◽

Price Changes ◽

Term Memory ◽

The Future ◽

Long Short Term Memory

Forex is the largest global financial market in the world. Traditionally, fundamental and technical analysis are strategies that the Forex traders often used. Nowadays, advanced computational technology, Artificial Intelligence (AI) has played a significant role in the financial domain. Various applications based on AI technologies particularly machine learning and deep learning have been constantly developed. As the historical data of the Forex are time-series data where the values from the past affect the values that will appear in the future. Several existing works from other domains of applications have proved that the Long-Short Term Memory (LSTM), which is a particular kind of deep learning that can be applied to modeling time series, provides better performance than traditional machine learning algorithms. In this paper, we aim to develop a powerful predictive model targeting to predicts the daily price changes of the currency pairwise in the Forex market using LSTM. Besides, we also conduct an extensive experiment with the intention to demonstrate the effect of various factors contributing to the performance of the model. The experimental results show that the optimized LSTM model accurately predicts the direction of the future price up to 61.25 percent.

Download Full-text

A Deep Learning Integrated Lee–Carter Model

Risks ◽

10.3390/risks7010033 ◽

2019 ◽

Vol 7 (1) ◽

pp. 33 ◽

Cited By ~ 11

Author(s):

Andrea Nigri ◽

Susanna Levantesi ◽

Mario Marino ◽

Salvatore Scognamiglio ◽

Francesca Perla

Keyword(s):

Deep Learning ◽

Short Term Memory ◽

Predictive Accuracy ◽

Mortality Rates ◽

Short Term ◽

Term Memory ◽

The Future ◽

Long Short Term Memory ◽

Carter Model ◽

Over Time

In the field of mortality, the Lee–Carter based approach can be considered the milestone to forecast mortality rates among stochastic models. We could define a “Lee–Carter model family” that embraces all developments of this model, including its first formulation (1992) that remains the benchmark for comparing the performance of future models. In the Lee–Carter model, the κ t parameter, describing the mortality trend over time, plays an important role about the future mortality behavior. The traditional ARIMA process usually used to model κ t shows evident limitations to describe the future mortality shape. Concerning forecasting phase, academics should approach a more plausible way in order to think a nonlinear shape of the projected mortality rates. Therefore, we propose an alternative approach the ARIMA processes based on a deep learning technique. More precisely, in order to catch the pattern of κ t series over time more accurately, we apply a Recurrent Neural Network with a Long Short-Term Memory architecture and integrate the Lee–Carter model to improve its predictive capacity. The proposed approach provides significant performance in terms of predictive accuracy and also allow for avoiding the time-chunks’ a priori selection. Indeed, it is a common practice among academics to delete the time in which the noise is overflowing or the data quality is insufficient. The strength of the Long Short-Term Memory network lies in its ability to treat this noise and adequately reproduce it into the forecasted trend, due to its own architecture enabling to take into account significant long-term patterns.

Download Full-text

Volatility Forecasting for High-Frequency Financial Data Based on Web Search Index and Deep Learning Model

Mathematics ◽

10.3390/math9040320 ◽

2021 ◽

Vol 9 (4) ◽

pp. 320

Author(s):

Bolin Lei ◽

Boyu Zhang ◽

Yuping Song

Keyword(s):

Deep Learning ◽

Index System ◽

High Frequency ◽

Short Term Memory ◽

Learning Model ◽

Financial Data ◽

Volatility Forecasting ◽

Investor Attention ◽

Search Index ◽

Deep Learning Model

The existing index system for volatility forecasting only focuses on asset return series or historical volatility, and the prediction model cannot effectively describe the highly complex and nonlinear characteristics of the stock market. In this study, we construct an investor attention factor through a Baidu search index of antecedent keywords, and then combine other trading information such as the trading volume, trend indicator, quote change rate, etc., as input indicators, and finally employ the deep learning model via temporal convolutional networks (TCN) to forecast the volatility under high-frequency financial data. We found that the prediction accuracy of the TCN model with investor attention is better than those of the TCN model without investor attention, the traditional econometric model as the generalized autoregressive conditional heteroscedasticity (GARCH), the heterogeneous autoregressive model of realized volatility (HAR-RV), autoregressive fractionally integrated moving average (ARFIMA) models, and the long short-term memory (LSTM) model with investor attention. Compared with the traditional econometric models, the multi-step prediction results for the TCN model remain robust. Our findings provide a more accurate and robust method for volatility forecasting for big data and enrich the index system of volatility forecasting.

Download Full-text

Method of Evaluating and Predicting Traffic State of Highway Network Based on Deep Learning

Journal of Advanced Transportation ◽

10.1155/2021/8878494 ◽

2021 ◽

Vol 2021 ◽

pp. 1-9

Author(s):

Jiayu Liu ◽

Xingju Wang ◽

Yanting Li ◽

Xuejian Kang ◽

Lu Gao

Keyword(s):

Deep Learning ◽

Prediction Model ◽

Network Traffic ◽

Short Term Memory ◽

Real Data ◽

Hebei Province ◽

Discriminant Model ◽

Traffic State ◽

Highway Network

The accurate evaluation and prediction of highway network traffic state can provide effective information for travelers and traffic managers. Based on the deep learning theory, this paper proposes an evaluation and prediction model of highway network traffic state, which consists of a Fuzzy C-means (FCM) algorithm-based traffic state partition model, a Long Short-Term Memory (LSTM) algorithm-based traffic state prediction model, and a K-Means algorithm-based traffic state discriminant model. The highway network in Hebei Province is employed as a case study to validate the model, where the traffic state of highway network is analyzed using both predicted data and real data. The dataset contains 536,823 pieces of data collected by 233 continuous observation stations in Hebei Province from September 5, 2016, to September 12, 2016. The analysis results show that the model proposed in this paper has a good performance on the evaluation and prediction of the traffic state of the highway network, which is consistent with the discriminant result using the real data.

Download Full-text

Detection of Financial Statement Fraud Using Deep Learning for Sustainable Development of Capital Markets under Information Asymmetry

Sustainability ◽

10.3390/su13179879 ◽

2021 ◽

Vol 13 (17) ◽

pp. 9879

Author(s):

Chyan-Long Jan

Keyword(s):

Deep Learning ◽

Information Asymmetry ◽

Capital Markets ◽

Short Term Memory ◽

Performance Indicator ◽

Financial Data ◽

Financial Statement ◽

Financial Statement Fraud ◽

Financial Reports ◽

Financial Status

Information asymmetry is everywhere in financial status, financial information, and financial reports due to agency problems and thus may seriously jeopardize the sustainability of corporate operations and the proper functioning of capital markets. In this era of big data and artificial intelligence, deep learning is being applied to many different domains. This study examines both the financial data and non-financial data of TWSE/TEPx listed companies in 2001–2019 by sampling a total of 153 companies, consisting of 51 companies reporting financial statement fraud and 102 companies not reporting financial statement fraud. Two powerful deep learning algorithms (i.e., recurrent neural network (RNN) and long short-term memory (LSTM)) are used to construct financial statement fraud detection models. The empirical results suggest that the LSTM model outperforms the RNN model in all performance indicators. The LSTM model exhibits accuracy as high as 94.88%, the most frequently used performance indicator.

Download Full-text

Spatio-temporal warping for myoelectric control: an oﬄine, feasibility study

Journal of Neural Engineering ◽

10.1088/1741-2552/ac387f ◽

2021 ◽

Author(s):

Milad Jabbari ◽

Rami Khushaba ◽

Kianoush Nazarpour

Keyword(s):

Deep Learning ◽

Feasibility Study ◽

Short Term Memory ◽

Myoelectric Control ◽

Classification Error ◽

Feature Extraction Method ◽

Spatio Temporal ◽

New Feature ◽

Prosthesis Control ◽

The Time Domain

Abstract Objective: The efficacy of an adopted feature extraction method directly affects the classification of the electromyographic (EMG) signals in myoelectric control applications. Most methods attempt to extract the dynamics of the multi-channel EMG signals in the time domain and on a channel-by-channel, or at best pairs of channels, basis. However, considering multi-channel information to build a similarity matrix has not been taken into account. Approach: Combining methods of long and short-term memory (LSTM) and dynamic temporal warping (DTW), we developed a new feature, called spatio-temporal warping (STW), for myoelectric signals. This method captures the spatio-temporal relationships of multi channels EMG signals. Main Results: Across four online databases, we show that in terms of average classification error and standard deviation values, the STW feature outperforms traditional features by 5% to 17%. In comparison to the more recent deep learning models, e.g. convolutional neural networks (CNN), STW outperformed by 5% to 18%. Also, STW showed enhanced performance when compared to the CNN+LSTM model by 2% to 14%. All differences were statistically significant with a large effect size. Significance: This feasibility study provides evidence supporting the hypothesis that spatio-temporal warping of the EMG signals can enhance the classification accuracy in an explainable way when compared to recent deep learning methods. Future work includes real-time implementation of the method and testing for prosthesis control.

Download Full-text

A Deep Learning based Arabic Script Recognition System: Benchmark on KHAT

The International Arab Journal of Information Technology ◽

10.34028/iajit/17/3/3 ◽

2020 ◽

Vol 17 (3) ◽

pp. 299-305 ◽

Cited By ~ 1

Author(s):

Riaz Ahmad ◽

Saeeda Naz ◽

Muhammad Afzal ◽

Sheikh Rashid ◽

Marcus Liwicki ◽

...

Keyword(s):

Deep Learning ◽

Character Recognition ◽

Data Augmentation ◽

Short Term Memory ◽

Recognition System ◽

Learning Approach ◽

Arabic Text ◽

Data Set ◽

Processing Step ◽

Handwritten Arabic

This paper presents a deep learning benchmark on a complex dataset known as KFUPM Handwritten Arabic TexT (KHATT). The KHATT data-set consists of complex patterns of handwritten Arabic text-lines. This paper contributes mainly in three aspects i.e., (1) pre-processing, (2) deep learning based approach, and (3) data-augmentation. The pre-processing step includes pruning of white extra spaces plus de-skewing the skewed text-lines. We deploy a deep learning approach based on Multi-Dimensional Long Short-Term Memory (MDLSTM) networks and Connectionist Temporal Classification (CTC). The MDLSTM has the advantage of scanning the Arabic text-lines in all directions (horizontal and vertical) to cover dots, diacritics, strokes and fine inflammation. The data-augmentation with a deep learning approach proves to achieve better and promising improvement in results by gaining 80.02% Character Recognition (CR) over 75.08% as baseline.

Download Full-text

Human Activity Recognition using Fourier Transform Inspired Deep Learning Combination Model

International Journal of Sensors Wireless Communications and Control ◽

10.2174/2210327908666180727123657 ◽

2019 ◽

Vol 9 (1) ◽

pp. 16-31

Author(s):

Kyungkoo Jun

Keyword(s):

Fourier Transform ◽

Deep Learning ◽

Short Term Memory ◽

Window Size ◽

Sensor Data ◽

Data Sets ◽

Data Set ◽

Proposed Model ◽

Testing Data ◽

Labeling Scheme

Background & Objective: This paper proposes a Fourier transform inspired method to classify human activities from time series sensor data. Methods: Our method begins by decomposing 1D input signal into 2D patterns, which is motivated by the Fourier conversion. The decomposition is helped by Long Short-Term Memory (LSTM) which captures the temporal dependency from the signal and then produces encoded sequences. The sequences, once arranged into the 2D array, can represent the fingerprints of the signals. The benefit of such transformation is that we can exploit the recent advances of the deep learning models for the image classification such as Convolutional Neural Network (CNN). Results: The proposed model, as a result, is the combination of LSTM and CNN. We evaluate the model over two data sets. For the first data set, which is more standardized than the other, our model outperforms previous works or at least equal. In the case of the second data set, we devise the schemes to generate training and testing data by changing the parameters of the window size, the sliding size, and the labeling scheme. Conclusion: The evaluation results show that the accuracy is over 95% for some cases. We also analyze the effect of the parameters on the performance.

Download Full-text