Abstract

The use of computers and the internet has spread rapidly over the course of the past few decades. Every day, more and more people are coming to rely heavily on the internet. When it comes to the field of information security, the subject of security is one that is becoming an increasingly important focus. It is vital to design a powerful intrusion detection system in order to prevent computer hackers and other intruders from effectively getting into computer networks or systems. This can be accomplished by: (IDS). The danger and attack detection capabilities of the computer system are built into the intrusion detection system. Abuse has occurred and can be used to identify invasions when there is a deviation between a preset pattern of intrusion and an observed pattern of intrusion. An intrusion detection system (IDS) is a piece of hardware (or software) that is used to generate reports for a Management Station as well as monitor network and/or system activities for unethical behaviour or policy violations. In the current study, an approach known as machine learning is suggested as a possible paradigm for the development of a network intrusion detection system. The results of the experiment show that the strategy that was suggested improves the capability of intrusion detection.

Introduction

Over the past few years, there has been an increase in the usage of computer systems to make the lives of consumers easier and more convenient. When people try to takeadvantage of the amazing capabilities and processing capacity of computer systems, however, security has been one of the most significant problems in the field of computer science. This is because attackers frequently try to break into systems and act maliciously, such as stealing vital information from a corporation, rendering the systems useless, or even destroying the systems. Internal attacks, such as pharming, distributed denial-of-service (DDoS), eavesdropping, and spear-phishing attempts, are often among the most difficult to identify of all well-known attacks. This is due to the fact that firewalls and intrusion detection systems (also known as IDSs) often guard against attacks from the outside. At this time, the majority of systems authenticate users by analysing a login pattern consisting of the user ID and password. As a result of this, we have proposed in this study a security solution that we have dubbed the Internal Intrusion Detection and Protection System (IIDPS) [1]. This solution recognises hostile or malicious behaviour carried out against a system at the System call level. IIDPS uses data mining and forensic profiling techniques in order to mine system call patterns, also known as SC-patterns, which are the longest system call sequences (SC-sequences) that have repeatedly appeared numerous times in a user’s log file for the user. SC-patterns can be used to identify malicious activity. The user’s computer usage history is used to compile the user’s forensic features, which are then defined as a SCpattern that commonly appears in the user’s own submitted SCsequences but is rarely utilised by other users. This information is gleaned from the user’s computer.

Conclusion

As a consequence of this, within this work, we suggest asecurity system that we refer to as the Internal Intrusion Detection and Protection System (IIDPS). This system is capable of identifying hostile behaviour that is aimed towardsa system at the SC level. The IIDPS minessystem call patterns, also known as SC-patterns, which are defined as the longest system call sequences (System Call-sequences) that have repeatedly appeared numerous times in a user’s log file for the user. These are the operations carried out by the user, such as sending a file, updating a file, or viewing a file, and they are validated by an administrator. The user’s computer usage history is used to compile the user’s forensic features, which are then defined as a SC-pattern that commonly appearsin the user’s own submitted SC-sequences but is rarely utilised by other users. This information is gleaned from the user’s computer.

1- Introduction

2- Related work

3- Material and methods

4- Experimental Results

5- Conclusion

References

Abstract

Arrhythmia is one of the major cause of deaths across the globe. Almost 17.9 million deaths are caused due to cardiovascular diseases. In order to reduce this much mortality rate, the cardiovascular disease should be properly identified and the proper treatment for the same should be immediately provided to the patients. In this study, a new ensemble based support vector machine (SVM) classifier was proposed to classify heartbeat into four classes from MIT-BIH arrhythmia database. The results were compared with other classifiers that are SVM, Random Forest (RF), K-Nearest Neighbours (KNN), and Long Short Term Memory network. The four features were extracted from the ECG signals that were used by the classifiers are Wavelets, high order statistics, R-R intervals and morphological features. An ensemble of SVMs obtained the best result with an overall accuracy of 94.4%.

Introduction

The prime cause of deaths across the world according to the report of WHO are cardiovascular diseases (CVDs): Annually, more people die from cardiac diseases than from any other disease. In 2016, approximately 18 million people died because of cardiac-related diseases, which reflects 31% of all the deaths globally. 85% of the deaths among this 31% are due to heart attack and stroke. An approximate of three-fourths of cardiac deaths take place in low-income and middle-income countries [1]. In 2015, 82% of the 17 million premature deaths due to non- communicable diseases are in low and middle-income countries, and the rest are caused by CVDs. The leading cause of CVDs is a long-term effect of cardiac arrhythmias. When the electrical signal, to the heart that coordinate heartbeats don’t work properly, Arrhythmias occur [2].

Nowadays, it is important for the detection of ultrasound images of breast tumors. In this paper, a new ultrasonic image feature extraction algorithm combining edge-based features and morphologic feature information is proposed, which has good effect on benign and malignant identification of breast tumors. This paper mainly studies three features (Sum of maximum curvature, Sum of maximum curvature and peak, Sum of maximum curvature and standard deviation) according to the shape histogram of ultrasound breast tumors from a local perspective. Based on the results of SVM classifier, it was found that the edgebased features have higher classification accuracy. The recognition system would perform better when morphologic features (Roughness, Regularity, Aspect ratio, Ellipticity, Roundness) were incorporated, compared with the control group whose input only with morphologic features. The results show that edge-based features can well describe breast tumors in ultrasound images, and have the potential to be used in breast ultrasound computer-aided design.

Introduction

The statistic report in 2017 shows that the average age of breast cancer patients in China is 48.7 [1]. Breast cancer has become a common disease among women in the current society [2]. Both the breast cancer’s morbidity and mortality are higher than that of other female malignant tumors. Clinical studies have shown that early detection and effective treatments can greatly improve the survival rate of female patients. However, there was no obvious symptoms in the initial when the patient got the cancer, which makes the detection difficult. Therefore, how to discover the lesion area of breast as early as possible so as to improve the cure rate of breast cancer has become a very important topic in the medical field. Ultrasound imaging is a convenient, low-cost, effective, realtime, non-radiation imaging tool, which has been widely used in clinical breast cancer detection [3,4]. In breast tumor diagnosis, the breast ultrasound computer-aided diagnosis (CAD) has been becoming more and more important. It performs better in image preprocessing, segmentation, feature extraction and selection, and tumor classification, including the objective evaluation ∗ Corresponding author. E-mail address: tongying@njit.edu.cn (Y. Tong). results, the classification accuracy, and the diagnostic sensitivity [5,6]. The extraction of different features is crucial in breast ultrasound CAD. Over the past years, research prevailingly concerned the morphologic feature extraction and texture feature extraction [7–9]. Texture features mainly reflect the surface properties of objects through pixels’ gray distribution and their surrounding spatial neighborhoods’ properties like the clarity, thickness and depth of the image texture.

Malware families typically evolve over a period of time. Differences between malware samples within a single family can originate from various code modifications designed to evade detection, or changes that are made to alter the functionality of the malware itself. Thus, malware samples from the same family from different time periods can exhibit significantly different behavior. In this research, we apply feature ranking—based on linear support vector machine (SVM) weights—to identify changes within malware families. We analyze numerous malware families over extended periods of time. Our goal is to demonstrate that we can detect evolutionary changes within malware families using an automated and quantifiable machine learning based technique.

Introduction

Malware can be defined as malicious software that is designed to cause disruption, deny activity, gather private data without user consent, allow unapproved access to system resources, and similar improper behavior (Aycock, 2006). Malware detection and prevention is a high priority for governments and businesses. Building effective countermeasures to malware threats is difficult due to the complexity of modern software and networked systems. Creators of malware can take advantage of weaknesses in security mechanisms of networks and end systems. Hackers and organized criminals frequently introduce new features to enable their malware to evade detection. In addition, it is highly likely that much—if not most—new malware is written based of existing code, rather than starting from scratch (Walenstein, Venable, Hayes, Thompson, & Lakhotia, 2007). Typically, in the software development process, new software is written on top of the existing software, and in this sense, malware is no different. Thus, malware writers are inclined to reuse existing malware code and release new variants of the same malware (Aycock, 2006). For these reasons, malware can be viewed as evolving over time.

The main aim of this study is to introduce a two-step method for damage identification in moment frame connections using a support vector machine (SVM) and differential evolution algorithm (DEA). In the first step, the possibility location of damage in connections is determined through SVM leading to reducing the dimension of the search space. Then, the accurate location and precise amount of damage in connections are determined in the second step via DEA with a high speed. In order to simulate damage in connections, a moment frame is modeled through semi-rigid beam to column connections and the analytical model is used to randomly generate structures with damaged connections as data. Then, SVM is trained and tested using this data, to facilitate natural frequencies are considered as input data and the characteristic of damage in beam to column connections are considered as output data of the network. Now, the possible location of the damage in connections can be determined using the SVM trained. The accurate location and severity of damage are determined by DEA based on the prediction of SVM in the first step. In order to assess the efficiency of the proposed method, two numerical examples are considered with different damage cases and considering noise. A comparative study is also made to judge the performance of the method with that of a work available in the literature. The outcome shows the high efficiency of the proposed method to identify the location and severity of the damage in moment frame connections.

Introduction

Occurrence of damage in structural systems such as buildings, bridges, oil platforms and so on is inevitable in their lifetime. There are many examples of damage in structures that have been led to an overall failure. In order to prevent from spreading the local damage to overall one, it is important to identify and repair damage by inspecting the current status of structures. Damage identification methods are categorized into destructive and non-destructive methods. The destructive methods are not a suitable method for most structures because of their cost and inefficiency, hence, researchers turned to non-destructive methods. One of the most important non-destructive identification methods is based on observing the change in structural responses such as dynamic and static responses. The changes in structures due to damage are shown better by dynamic responses, made the dynamic based methods more popular. The damage identification in structures should be in some way that the location and severity of damage in structures are accurately determined. Over the last few years, various methods have been proposed to identify damage in structural members, however, damage identification in connections has been less studied. This issue in earthquake-zone areas that a localized damage in connections may be led to an overall failure of structure, increases the importance of damage identification in connections. In 2001, a research was carried out by Yun et al. for estimating the joint damage of a steel structure from modal data using a neural network technique.

Literature review

Methodology

Data analysis and results

Conclusion

References

Abstract

This study examines the predictability of real-time news data on investors’ buying behaviour in the futures market, using supervised sentiment analysis. Market sentiment or traders’ buying behaviour is captured at the bid-ask stage of price formation using the net buying pressure (NBP). Any significant change in NBP patterns defines an “interesting market event”. Real-time news headlines are automatically labelled using interesting market events, assuming a lag between the market information and its impact on buying behaviour. News was found to have an impact on the market buying behaviour of the S&P NIFTY index futures with an optimal lag of 5 minutes. Manual labelling of the news data validated this empirical finding.

Literature review

Prior studies have established the predictability of the impact of news on market sentiment in the spot market context. Supervised sentiment analysis is a text classification task where the impact of textual market information on financial markets is learnt by using labelled training data. The creation of the training data involves labelling the news instances according to their impact on the markets. The news instances can be labelled manually according to the news discourse (Davis et al., 2006; Bozic & Seese, 2011) or automatically based on the corresponding market trends (Mittermayer & Knolmayer, 2006). Labelling news instances manually is a precise but labour-intensive task; hence, this method is not suitable for high-frequency news analytics. Automatically aligning news instances with the corresponding market trends makes for one of the most challenging tasks in sentiment analysis. Yoo et al. (2005), Mittermayer & Knolmayer (2006), Nikfarjam et al. (2010), and Nassirtoussi et al. (2014) reviewed studies that have used the supervised sentiment analysis approach. A brief review of some of the major concerns that our study deals with, follows. News-trend alignment is one of the most important parts of a supervised sentiment analysis model for testing and contextualising market efficiency. The accuracy of the newstrend alignment procedure constrains the efficacy of the supervised sentiment analysis.

1- Introduction

2- SVM method

3- Portfolio selection model: MV method

4- The experiment

5- Results

6- Conclusions

References

Abstract

Forecasting stock returns is an exacting prospect in the context of financial time series. This study proposes a unique decision-making model for day trading investments on the stock market. In this regard, the model was developed using a fusion approach of a classifier based on machine learning, with the support vector machine (SVM) method, and the mean-variance (MV) method for portfolio selection. The model’s experimental evaluation was based on assets from the São Paulo Stock Exchange Index (Ibovespa). Monthly rolling windows were used to choose the best-performing parameter sets (the in-sample phase) and testing (the out-of-sample phase). The monthly windows were composed of daily rolling windows, with new training of the classifying algorithm and portfolio optimization. A total of 81 parameter arrangements were formulated. To compare the proposed model’s performance, two other models were suggested: (i) SVM + 1/N, which maintained the process of classifying the trends of the assets that reached a certain target of gain and which invested equally in all assets that had positive signals in their classifications, and (ii) Random + MV, which also maintained the selection of those assets with a tendency to reach a certain target of gain, but where the selection was randomly defined. Then, the portfolio’s composition was determined using the MV method. Together, the alternative models registered 36 parameter variations. In addition to these two models, the results were also compared with the Ibovespa’s performance. The experiments were formulated using historical data for 3716 trading days for the out-of-sample analysis. Simulations were conducted without including transaction costs and also with the inclusion of a proportion of such costs. We specifically analyzed the effect of brokerage costs on buying and selling stocks on the Brazilian market. This study also evaluated the classifier’s performance, portfolios’ cardinality, and models’ returns and risks. The proposed main model showed significant results, although demand for trading value can be a limiting factor for its implementation. Nonetheless, this study extends the theoretical application of machine learning and offers a potentially practical approach to anticipating stock prices.

Introduction

Predicting stock returns is considered to be one of the most challenging tasks when dealing with financial time series because the stock market is dynamic, complex, evolutionary, nonlinear, nebulous, nonparametric, and chaotic by nature. Additionally, the stock market is extremely sensitive to political factors, microeconomic and macroeconomic conditions, and investors’ expectations and insecurities (Ballings, Van den Poel, Hespeels, & Gryp, 2015; Kara, Boyacioglu, & Baykan, 2011; Tan, Quek, & Ng, 2007). According to mainstream financial theory, predicting financial asset prices is impossible. The efficient market hypothesis (EMH), which is the literature’s main theoretical pillar, suggests that the task of predicting future prices based on financial assets’ past behavior cannot achieve abnormal returns. The reason is that the distribution function of a financial time series denotes a Brownian motion, which has random, independent, and Gaussian distribution characteristics. However, some studies reject the EMH, arguing that the stock market is not actually established at random and that financial time series have long-term memory. For example, recent studies dispute the EMH in the context of different markets and periods. Cervelló-Royo, Guijarro, and Michniuk (2015) showed empirical evidence that compromises the premise of the EMH. The authors researched the intraday markets of the American Dow Jones Industrial Average (DJIA) index, the German Deutscher Aktien 30 Index (DAX), and the British Financial Times Stock Exchange (FTSE) index from 2000 to 2013. By using a methodology based on flag patterns, they tested 96 different configurations. The results were very profitable for all the markets. Chourmouziadis and Chatzoglou (2016) researched daily data from the Athens Stock Exchange from 1996 to 2012. They combined technical analysis and fuzzy logic. The profitability results of the proposed methodology were surprisingly higher than the baselines. Kim and Enke (2016) studied the intraday data of the Korea Composite Stock Price Index (KOSPI) 200 from 2007 to 2014. They formulated a set of rules based on technical analysis combined with genetic algorithms and reported abnormal profits. Chen and Chen (2016) proposed a model based on flag patterns for recognizing bullish reversal patterns. They tested the strategy with the National Association of Securities Dealers Automated Quotation System (NASDAQ) from 1989 to 2004 and with the Taiwan Capitalization Weighted Stock Index (TAIEX) from 1990 to 2004. The results showed high levels of profitability. Kampouridis and Otero (2017) studied the intraday foreign exchange market, analyzing a 10-month period between 2013 and 2014. They proposed a model that combined technical indicators, physical time scales, and genetic algorithms. The results guaranteed higher returns than those of the baselines.

I- Introduction

II- Classification of plant diseases

III- Classification algorithms

IV- Conclusion

References

Abstract

Plants are considered to be important as they are the source of energy supply to mankind. Plant diseases can affect the leaf any time between sowing and harvesting which leads to huge loss on the production of crop and economical value of market. Therefore, leaf disease detection plays a vital role in agricultural field. However, it requires huge manpower, more processing time and extensive knowledge about plant diseases. Hence, machine learning is applied to detect diseases in plant leaves as it analyzes the data from different aspects, and classifies it into one of the predefined set of classes. The morphological features and properties like color, intensity and dimensions of the plant leaves are taken into consideration for classification. This paper presents an overview on various types of plant diseases and different classification techniques in machine learning that are used for identifying diseases in different plant leaves.

INTRODUCTION

INDIA is a fast developing country and agriculture is the back bone for the country’s development in its early stage. However, agricultural field faces lots of hurdles including huge loss in the crop production. Plant leaf diseases are one of the important reasons for the loss in the production and plant leaf disease identification is also very difficult in agriculture field. Naked eye method is a traditional method of identifying the diseases which involves huge man power, inaccurate, time consuming and not applicable for larger fields. In addition, it is very expensive as it requires continuous monitoring by the experts. Hence, machine learning; a reliable prediction methodology is used for detecting various diseases of plant leaves caused by fungus, bacteria and virus. However, disease prediction using classification algorithms appears to be a difficult task as the accuracy varies for different input data. In this paper, several research contributions related to various plant leaf diseases detection using different classification algorithms are reviewed and compared.

Air pollution can lead to a wide range of hazards and can affect most organisms on Earth. Therefore, managing and controlling air pollution has become a top priority for many countries. An effective short-term atmospheric pollutant concentration forecasting (SAPCF) can mitigate the negative effects of atmospheric pollution. In this paper, we propose a new hybrid forecasting model for SAPCF. Firstly, we analyse the influential factors of pollutants to obtain the optimal combination of input variables. Secondly, we use a clustering algorithm to enhance the regularity of our modelling data. Thirdly, we build a particle swarm optimisation (PSO)–support vector machine (SVM) hybrid model called PSO–SVM and perform a case study in Temple of Heaven, Beijing to test its forecasting accuracy and validate its performance against three contrastive models. The first model inputs all possible variables in equal weight without influence factor analysis. The second model integrates the same input variables used in the proposed model without clustering. The third model inputs these same variables with genetic-algorithm optimised SVM parameters. The comparison amongst these models demonstrates the superior performance of our proposed hybrid model. We further verify the forecasting results of our hybrid model by conducting statistical tests.

Introduction

As one of the four elements of life, air plays a major role in maintaining the ecosystem. However, human activities have seriously aggravated the degree of air pollution, thereby prompting researchers to conduct pollution analysis (Bollen, 2015) and predict pollutant concentrations (Wu et al., 2018). Given its important role in formulating effective precautionary measures, atmospheric pollutant concentration forecasting (APCF) has received much research attention (Bai et al., 2018). Previous studies have classified APCF into short-term APCF (SAPCF) (Niu et al., 2016), medium and long term APCF. Medium and long term APCF forecasts the concentration of pollutants over a relatively long period, usually ranging from months to years (Nebenzal and Fishbain, 2018), and is mainly used for planning the distribution of industrial sites or residential areas. Meanwhile, SAPCF reveals the vital status in many basic operations and is often used in planning abatement actions and transportation networks in advance (Li and Tao, 2018; Xie et al., 2018; Zhai and Chen, 2018). SAPCF can also help governments save time in responding to pollution-related problems (Y.F. Wang et al., 2018) and help individuals prevent exposure to pollutants (Soh et al., 2018). Therefore, an accurate SAPCF is of great significance at the social and individual levels. Researchers have proposed numerous methods for SAPCF in recent years as will be discussed in Section 1.1.

Multiconlitron is a general theoretical framework for constructing piecewise linear classifier. However, it contains a relatively large number of linear functions, resulting in complicated model structure and poor generalization ability. Learning to prune redundant or excessive components may be a very necessary progression. We propose a novel greedy method, i.e., greedy support multiconlitron algorithm (GreSMA) to simplify the multiconlitron. In GreSMA, a procedure of greedy selection is first used. It generates the initial linear boundaries, each of which can separate maximum number of training samples under the current iteration. In this way, a minimal set of decision functions is established. In the second stage of GreSMA, a procedure of boundary adjustment is designed to retrain the classification boundary between convex hulls of local subsets, instead of individual samples. Thus, the adjusted boundary will fit the data more closely. Experiments on both synthetic and real-world datasets show that GreSMA can produce minimal multiconlitron with better performance. It meets the criteria of ‘‘Occam’s razor’’, since simpler model can help prevent over-fitting and improve the generalization ability. More significantly, the proposed method does not contain parameters that depend on the datasets or make assumptions of the underlying statistical distributions of the samples. Therefore, it should be regarded as an attractive advancement of piecewise linear learning in the general framework of multiconlitron.

Introduction

In pattern recognition, piecewise linear classifier (PLC) is effective when a statistical model cannot express the underlying distribution of samples [1]. It approximates the true classification boundary by a combination of hyperplanes. Since each piece is linear, a PLC is very simple to implement with requirement of low memory usage. Therefore, it has the potential to be applied to the scenarios of small reconnaissance robots, intelligent cameras, embedded and real-time systems, and portable devices [2]. Despite the simplicity in implementation, constructing a PLC usually requires complex computational procedure [3]. In general, there are two criteria that need to be considered, i.e., selecting appropriate number of hyperplanes and minimizing the error of classification. Under their guidance, many methods have been presented to synthesize PLCs over the last few decades. Hierarchical partitioning is one of the common ways. In 1996, Chai et al. [4] achieved a binary tree structure with genetic algorithm to design a PLC in the sense of maximum impurity reduction. For simplifying the construction of a decision-tree PLC, in 2006 Kostin [2] developed and implemented a simple and fast multi-class PLC with acceptable classification accuracies, based on tree division of subregion centroids. In 2016, Wang et al. [5] proposed hierarchical mixing linear support vector machines (SVMs) for nonlinear classification, which can be seen as special form of a PLC. Furthermore, Ozkan et al. [6] designed a highly dynamical self-organizing decision tree structure for mitigating overtraining issues.

I. INTRODUCTION

II. METHODOLOGY

III. EXPERIMENTAL RESULTS AND DISCUSS

IV. CONCLUSION

REFERENCES

This paper presents a Synthetic Aperture Radar (SAR) image classification algorithm based on multi-feature using Fruit Fly Optimization Algorithm (FOA) and Least Square Support Vector Machine (LS-SVM). First, pixel-based information derived from three elements of coherency matrix, six parameters obtained by H/α/A decomposition and Freeman decomposition techniques, and three polarimetric parameters including the total receive power (SPAN), pedestal height, and Radar Vegetation Index (RVI), as well as region-based information derived from eight texture parameters obtained by Grey Level Co-occurrence Matrix (GLCM) are combined to use as the features of land cover. Second, Kernel Principal Component Analysis (KPCA) is used to reduce the dimensionality of the multi-feature data derived from the integration of the pixel-based and region-based information. Third, LS-SVM is used as the classifier in this study due to its fast solving speed and desirable classification capability. Since the input parameters of LS-SVM significantly affect the classification performance, we employ FOA to obtain the optimized input parameters. Finally, the experiments on two fully polarimetric SAR images of various crops with a limited number of samples are implemented by the proposed method and other commonly used methods, respectively. The results show that the proposed method can attain better classification performances compared with other methods.

INTRODUCTION

Land cover information is important for land development and management. Land cover classification is also the first step in remote sensing of vital global parameters such as soil moisture. Remotely sensing data obtained from various sensors provides an economical way to characterize land cover information. Optical remote sensing is an effective approach but by itself is limited by weather conditions. Synthetic Aperture Radar (SAR), which can obtain information under different weather conditions, is therefore used for acquiring land cover information in various regions. Significant research aiming at land cover classification has been reported by many researchers. In the early years, most studies were developed based on single-polarization data [1], [2]. Since single-polarization data does not contain all the polarization information of ground objects, such methods were most likely to create confusion among similar ground objects and thereby were only suited for coarse classification [3]–[5]. With the rapid development of the SAR techniques, many methods utilizing multi-polarization or full-polarization data were explored for attaining a better classification [6]–[8]. The critical procedure for these methods is polarimetric decomposition, which provides a way to obtain the physical features of natural media. Many polarimetric decomposition methods have been explored by many researchers [9]–[14].

1- Introduction

2- Literature review

3- Data collection and processing

4- Model development

5- Experimental evaluation

6- Conclusions and future work

References

Abstract

This work contributes to this ongoing debate on the nature and the characteristics of propagation channels of crash events in international stock markets. Specifically, we investigate transmission mechanisms across stock markets along with effects from bond and currency markets. Our approach comprises a solid forecasting mechanism of the probability of a stock market crash event in various time frames. The developed approach combines different machine learning algorithms which are presented with daily stock, bond and currency data from 39 countries that cover a large spectrum of economies. Specifically, we leverage the merits of a series of techniques including Classification Trees, Support Vector Machines, Random Forests, Neural Networks, Extreme Gradient Boosting, and Deep Neural Networks. To the best of our knowledge, this is the first time that Deep Learning and Boosting approaches are considered in the literature as a means of predicting stock market crisis episodes. The independent variables included in our data contain information regarding both the two fundamental linkage channels through which financial contagion can be initiated: returns and volatility. We apply a suite of machine learning algorithms for selecting the most relevant variables out of a large set of proposed ones. Finally, we employ bootstrap sampling for adjusting the imbalanced nature of the available fitting dataset. Our experimental results provide strong evidence that stock market crises tend to exhibit persistence. We also find significant evidence of interdependence and cross-contagion effects among stock, bond and currency markets. Finally, we show that the use of Deep Neural Networks significantly increases the classification accuracy, while offering a robust way to create a global systemic early warning tool that is more efficient and risk-sensitive than the currently established ones. Thus, central banks may use these tools to early adjust their monetary policy, so as to ensure financial stability.

Introduction

A global financial crisis can emerge from a series of local or/and regional market shocks, which evolve into a worldwide economic crisis due to the interconnectedness of the financial markets. For example, the Asian crisis in 1997 initially originated in Thailand; subsequently, it propagated to other Asian countries, and eventually made it to the financial markets of the United States of America and Europe (Kaminsky et al., 1998). In other cases, a crisis may start from a single economy whose size is large enough to generate turbulence in other countries. This is the case, for instance, with the subprime crisis that started in the United States and evolved into a sovereign debt crisis in several European countries. The observation that an economic crisis is manifested by a subsequent recession (Bluedorn et al.. (2013), Barro and Ursua (2009), Estrella and Mishkin (1998), Farmer (2012)) renders reliable Early Warning Systems (EWSs) valuable tools for policymakers, in their effort to curtail contagion risk and, in extreme cases, even preempt a global economic crisis. An EWS must be capable of producing clear signals as to whether an economic crisis is imminent, complementing the expert judgment of policymakers. Hence, EWS systems facilitate policy makers in unveiling vulnerabilities of the economy and taking precautionary actions to diminish the risks that can trigger a crisis. Certainly, there is always a trade-off between developing EWSs that are capable of predicting a lot of alarms for an imminent crisis, at the expense of some of them being wrong (false-alarms), and EWSs that predict rather too few signals of impending crises, at the expense of missing a major crisis event. Optimally, an EWS should let no crisis events go unnoticed, while minimizing the number of generated false-alarms. It goes without saying that the cost of not signaling a global crisis is significantly higher than that of an incorrect alarm. At the same time, the incorporation of the probability of a worldwide crisis in decisions related to asset al.location (Kole et al.., 2006) can substantially benefit investors. Indeed, this is the case since a global crisis significantly curtails diversification benefits, as worldwide markets move in the same direction. In addition, any hedging strategies may become ineffective (Ibragimov and Walden, 2007) due to the structural changes in the observed correlations among asset classes. Indeed, during periods of high volatility in bear markets, correlations increase across assets (Longin and Solnik, 2001). Thus, as the markets cannot quickly correct any disruptions in their function, it becomes even more imperative for regulators to intervene so as to restore financial stability.

In view of the two problems of the SVM algorithm in processing large data, the paper proposed a weighted Euclidean distance, radial integral kernel function SVM and dimensionality reduction algorithm for large data packet classification. The SVM cannot handle multi classification and time of building model is long. The algorithm solved these problems. The improved algorithm reconstructs the data feature space, makes the boundary of different data samples clearer, shortens the modeling time, and improves the accuracy of classification. The proposed method verified the feasibility and effectiveness with experiments. The experimental results show that the improved algorithm can achieve better results when multi-duplicated samples and large data capacity are used for multi classification.

INTRODUCTION

The rapid development of network technology makes a huge amount of data every day. The rapid and accurate classification of the vast amounts of data collected is necessary to extract comprehensible knowledge. According to forecasting by market research firm IDC, global data will exceed 40ZB by 2020[1]. Many industries have provided storage systems with capacity ranging from tens of gigabytes to hundreds of terabytes, or even petabytes. But nearly 60% of the data is repeated, which not only increases data storage, processing time, but also leads to higher and higher costs of data analysis and classification. The efficient and accurate classification algorithm is one of the hot issues in current industry research. There are some common classification algorithms. For example, K-Nearest Neighbor ǃ Native BayesǃNeural NetǃSupport Vector Machine and Linear Least Square Fit and so on[2]. Support vector machine (SVM) algorithm is a kind of machine learning method based on VC dimension theory in statistical learning theory and structural risk minimum principle. It has excellent data classification and regression processing ability[3]. The support vector method was first proposed by Vapnik to solve the problem of pattern recognition. It selects a set of characteristic subsets from the training samples, so that the classification of the characteristic subset is equivalent to the division of the whole dataset. The characteristic subset is called the support vector (SV). Due to its excellent learning ability, the application scope is very wide. For example, intrusion detection, facial expression classification, Time series prediction, speech recognition, signal processing, Gene detection, text classification, font recognition, Fault diagnosis, chemical analysis, image recognition and other fields. SVM algorithm has some obvious advantages in solving classification problems. It has a shorter forecast time. The global optimal solution can guarantee the accuracy of the target detection classifier in the classification. But there are some disadvantages, such as the detection model is established for a long time. Time complexity and space complexity increase linearly with the increase of data when processing large scale data. The data objects are often large data sets in the emerging fields of data mining, document classification and multimedia indexing. The number of attributes and the number of records are very large resulting in poor execution of the processing algorithm[4]. The classifier is only determined based on support vector machine by support vector. The complexity of the classifier is not related to the number of training samples. It only has to do with the number of support vectors[5]. In the paper, propose a weighted Euclidean distance, the radial product kernel function and the decreasing dimension packet support vector machine method, reduces the data dimension, remove redundant feature attributes and duplicate data. A classification model with better generalization ability is obtained by using less support vectors. Reducing storage and processing of data resources, speed up the classification model established time, solve big data classification problems.