Exploratory Review of Applications of Machine Learning in Finance Sector Sandip Rakshit , Nyior Clement, and Narasimha Rao Vajjhala Abstract The finance sector is one of the key pillars of any nation’s economy. However, with the emergence of big data and rapid advancements in technology, the finance sector is processing significant amounts of heterogenous data. Institutions in the finance sector are increasingly using machine learning algorithms and techniques to process these heterogenous data. This exploratory review provides an in-depth look at the machine learning applications in the finance sector. The state-of-the-art machine learning applications in the finance sector were reviewed in this exploratory study. The primary research question addressed in this study was to explore the machine learning algorithms and techniques applied to the applications in the finance sector. Various machine learning algorithms and techniques used in finance sector were broadly discussed in this study. This study also provides some suggestions about how machine learning can maximize productivity in the finance sector. Keywords Machine learning · Supervised learning · Unsupervised learning · Finance · Security · Algorithmic trading · Artificial intelligence · Data science 1 Introduction Machine learning (ML) algorithms and techniques in conjunction with other technologies can help process and use large volumes of heterogeneous data [1, 2]. Financial sector, in particular, can benefit significantly from the use of machine learning algorithms and techniques. The financial sector is one of the key pillars S. Rakshit · N. Clement American University of Nigeria, Yola, Adamawa, Nigeria e-mail: sandip.rakshit@aun.edu.ng N. Clement e-mail: nyior.clement@aun.edu.ng N. R. Vajjhala (B) University of New York Tirana, Kodra e Diellit, Tirana, Albania e-mail: narasimharao@unyt.edu.al © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 S. Borah et al. (eds.), Advances in Data Science and Management, Lecture Notes on Data Engineering and Communications Technologies 86, https://doi.org/10.1007/978-981-16-5685-9_12 119 120 S. Rakshit et al. of the economy of any nation. Some might even argue that an economy’s health relies majorly on its financial sector. Thus, there is a linear relationship between an economy’s health and its financial sector’s strength. While most people limit their understanding of what the financial sector is to exchanges made at the marketplace, there is much more to the financial sector than that. The financial industry is itself made up of smaller industries/sectors. The financial sector’s sub-sectors are usually financial institutions and firms like banks, real estate firms, insurance, and investment companies, providing financial services to their commercial and retail customers. Some of the services offered by the institutions in this sector are but not limited to providing loans to businesses for expansion, provision of mortgages to homeowners, insuring lives and assets, and building up savings for retirement. The value created by this industry increases as the interest rate drops. That is because a reduction in interest rate attracts more investment. 1.1 An Overview of Machine Learning The technology industry’s focus has been drifting toward building intelligent machines, machines that can have the ability to automate repetitive tasks with little or no human involvement. Machine learning is a methodology relying on a premise that machines should learn from the provided data as well as experiences relying primarily on data mining and learning algorithms [1]. Machine learning, a subset of artificial intelligence, uses general methodology to solve a number of problems. Some of the areas where machine learning algorithms have been successful, include mail filtering, optical character recognition, and computer vision [3]. Even though machine learning has been here for decades, its relevance rose due to the immense availability of data and the emergence of powerful computers at a lesser cost. In machine learning, an algorithm is usually trained with a training data set, and a trained algorithm gives rise to a model. The quality of prediction of a machine learning algorithm depends on the quality of the data used in training the algorithm [4]. Machine language algorithms can be classified into three categories to compare the classifiers, namely individual classifiers, homogeneous, and heterogeneous ensembles. The summary of machine learning techniques is provided in Fig. 1. Individual classifiers rely on a single machine learning algorithm [3]. Some of the techniques including decision tress, support vector machines, and neural networks are examples of individual classifiers. Model development and forecast combination are key steps in the ensemble method. Ensemble methods include both homogeneous and heterogeneous ensemble classifiers. Homogeneous classifiers such as bagging and boosting can be useful in increasing the accuracy of prediction of the classifiers [3]. Exploratory Review of Applications of Machine Learning … 121 Fig. 1 Summary of machine learning methods 1.2 Classification of Machine Learning Algorithms The application domain of machine learning algorithms can be divided into several categories, including supervised, unsupervised, semi-supervised, and reinforcement learning techniques [1, 3]. Supervised learning techniques and algorithms should ideally be applied in situations where both the predictors and responses are available. In this learning style, machine learning algorithms are trained with labelled data. Supervised learning algorithms help deduce a function from a labelled training that includes the instances as well as the anticipated outcome for each value of the instances [5]. For instance, an algorithm is provided with an image as input. The expected output is the content in the image. The algorithm learns by comparing its output with the expected output. In this example, the data is the image, and the label is the name of the provided image. There are two major supervised learning problems, namely: classification and regression problems. The way algorithms are trained and are formally called their learning style or machine learning method. Unsupervised learning techniques and algorithms are suitable for application in situations where only predictors, such as independent or exploratory variables are available [3]. Unsupervised learning techniques help in extracting the most distinct features of the data. Unsupervised learning algorithms help deduce a function to specify the unseen structure of the unlabelled data [5]. In this learning style, an algorithm is trained with unlabelled data. The algorithm explores the unlabelled data and finds some structure within the data that it could use to generalize. Semi-supervised learning algorithms are trained with both labelled and unlabelled data. It is usually more of the labelled data and less of unlabelled data. Semisupervised learning techniques help deduce a function grounded on smaller amount of labelled training dataset and a considerably larger amount of unlabelled dataset [5]. 122 S. Rakshit et al. The purpose of semi-supervised learning is to make optimal use of the large unlabelled samples [6]. Semi-supervised learning includes semi-supervised clustering and semi-supervised classification [6]. Reinforcement learning techniques are often used for robotics, gaming, and navigation. With reinforcement learning, the algorithm discovers through trial and error which actions yield the greatest rewards. This type of learning has three primary components: the agent (the learner or decision-maker), the environment (everything the agent interacts with), and actions (what the agent can do). The objective is for the agent to choose actions that maximize the expected reward over a given amount of time. 2 Applications of Machine Learning in Finance Sector The use of machine learning algorithms and techniques in dealing with financial data has several advantages over the traditional techniques. Machine learning techniques can automatically identify hidden features in the financial data apart from processing data with nonlinear characteristics [7]. With the rapid growth in finance, it has become necessary to automate some processes and secure existing processes. Several applications in the financial sector, including credit risk management, portfolio management, automatic trading, and fraud detection use machine learning algorithms and techniques [3]. Because machine learning primarily concerns itself with parsing and learning from data and big data is one of the important things the finance sector has, this science can be applied in several finance domains. Machine learning has several applications in finance. Some of the machine learning methods frequently used by the finance sector, include k-NN, decision trees, support vector machines, neural networks, and boosting [3]. This paper seeks to answer the following research questions: (a) (b) What machine learning algorithms and techniques are applied to the applications in the financial sector? What are the practical implications of using machine learning algorithms in the financial sector? 3 Analysis of Extant Literature Portfolio management is one of the commonest applications of machine learning in finance. They provide an alternative to the traditional and manual consulting individuals do with financial experts concerning their investment options. More formally put, we could say Robo—Advisors are digital platforms or online applications that automate investment processes. They provide financial guidance and services to clients. They achieve this through the use of algorithms and statistics. One of the Exploratory Review of Applications of Machine Learning … 123 Fig. 2 Conceptual model for algorithmic trading significant advantages of Robo—Advisors lies in the fact that they greatly simplify the entire investment process that could have been otherwise a daunting task. Algorithmic trading is another key area in the finance sector employing machine learning algorithms and techniques. Before the technological era, trading was purely a paper-based activity. Algorithmic trading has four major components: data component, model component, execution component, and monitor component. Figure 2 depicts these components. Buyers and sellers had to be physically present to purchase or sell goods and services, and certificates for these goods were offered at the end of each transaction. However, after some time, electronic certificates replaced the existing physical certificates, and trading was faster. That notwithstanding, trading in those days was inefficient by far. As a result, there was a need to digitize the trading process and make it more efficient and profitable, bringing about algorithmic trading. Algorithmic trading, as its name implies, refers to the process of using computer programs (algorithms) that instructs the computer to buy and sell stocks on one’s behalf when certain market conditions are met. These algorithms take advantage of the computing resources available to them and trade at a speed and frequency that is impossible to match by human traders. Fraud detection is also another area within the finance sector using machine learning techniques extensively. Fraud is the deliberate misrepresentation of a material fact with the intent of misleading a person or entity into acting upon it, resulting in the harm of the person or entity acting upon it [8]. With the recent emergence of machine learning, this technology has found numerous applications in this domain concerning this use case. Machine learning has been applied in finance to reduce the high occurrences of “false positives” (a situation whereby a financial institution or merchant declines a valid transaction request based on suspecting it to be fraudulent). Several machine learning algorithms including logistic regression, support vector machines, decision trees, and random forests were used in the domain of 124 S. Rakshit et al. credit card fraud detection [9]. Prior research indicates that random forest is one of the simplest and most suitable algorithms in the case of credit card fraud detection. Support vector machines gave the best result when applied to dataset with minimal fraud rates while giving moderate accuracy as the fraud rates increased [9]. Even though chatbots have been in existence for quite some time, machine learning-based chatbots brought about a new chatbot experience in finance. These new breeds of chatbots greatly improved human–computer interaction and, by extension, the end-user experience. This is made possible due to these chatbots’ natural language processing capability and their ability to learn from past experiences. As a result of these immense capabilities that these chatbots possess, they can adapt to every customer and his/her behavioral changes. These chatbots can do this because they parse tons of customer finance queries. Money is also one of the common occurrences in finance. However, with the emergence of machine learning, financial institutions seek ways to use machine learning models to reduce these events. These models will be trained to spot signs of money laundering in these financial transactions. Khac et al. [10] present a solution developed as a tool for identifying and analyzing money laundering using real transaction datasets. The authors used a three-level data mining framework comprising of various components, namely the pre-processing of data, data mining, and knowledge management. Lin-Tao et al. [11] proposed a method for increasing the detection rate of suspicious financial transactions and money laundering attempts with low false positives. Deng et al. [12] proposed active learning through sequential design method for prioritization to improve the process of money laundering detection. Salehi et al. [13] suggest unsupervised machine learning algorithms for effective money laundering detection given the crime’s dynamic nature. 4 Conclusion and Future Research Directions Even though machine learning has a plethora of applications in finance, significant technological advancements are expected in the next decade. Numerous improvements could be made in the application of machine learning in the financial sector. For example, sentiment analysis could be heavily applied in finance to deduce people’s emotions at any given point in time. This could, in turn, be used to enhance machine– human interaction. Furthermore, finance is one sector where security is paramount. Everyday existing security protocols are undergoing improvements, and new ones are being implemented to ensure users’ financial data are being properly secured. However, passwords and user names remain the primary user authentication mechanism, even in the financial sector for now. Given that user security in this sector is a high-stakes game, there is the need to architect a more stringent mechanism for ensuring user security. Machine learning can be used to enhance user security is through facial recognition, voice recognition, and biometric data. Despite the positive applications of machine learning in finance, it will be wrong to overlook the negative impact this technology could have on this economic sector. However, the Exploratory Review of Applications of Machine Learning … 125 widespread adoption of AI could introduce new systemic and security risks in the financial system. The early big movers offer their AI applications (that includes machine learning) as a “service” to their competitors, attracting users to accelerate their system’s learning and turning cost centers into profit centers. As this trend widens, the financial system may face new risks. In conclusion, even though machine learning has an increasingly wide range of finance applications, these applications would grow in the forthcoming years. It is as a result of this that finance institutions are now investing in this technology. Their investments are indeed bringing them a lot of benefits. Some of these benefits include a drastic reduction in operational costs. Because of this reduction in operating expenses, there tends to be an equal increase in revenue. Machine learning in finance also leads to an increase in customer loyalty due to a better user experience. In the meantime, machine learning algorithms in finance offer investment advice, reduce fraud, and trade on behalf of financial institutions. While busy with all these tasks, these algorithms are always learning and getting smarter day by day and bringing the world closer to completely automated financial processes. References 1. Chagas BNR et al (2020) A literature review of the current applications of machine learning and their practical implications. Web Intell (2405–6456), 18(1):69–83 2. Agarwal A, Jayant A (2019) Machine learning and natural language processing in supply chain management: a comprehensive review and future research directions. Int J Business Insights Transf 13(1):3–19 3. Teng H-W, Lee M (2019) Estimation procedures of using five alternative machine learning methods for predicting credit card default. Rev Pac Basin Financ Mark Policies 22(03):1950021 4. Moruff OA, Maruf AO, Tosho A (2020) Performance analysis of selected machine learning algorithms for the classification of phishing URLs. J Comput Sci Control Syst 13(2):16–19 5. Ahmad A et al (2020) A systematic literature review on using machine learning algorithms for software requirements identification on stack overflow. Sec Commun Netw p 1–19 6. Zhao J et al (2018) Safe semi-supervised classification algorithm combined with active learning sampling strategy. J Intell Fuzzy Syst 35(4):4001–4010 7. Chen Y et al (2020) Financial trading strategy system based on machine learning. In: Mathematical problems in engineering, pp 1–13 8. Ogbodo UK, Mieseigha EG (2013) The economic implications of money laundering in Nigeria. Int J Acad Res Accounting Finance Manage Sci 3(4):170–184 9. Sravya K et al (2020) Credit card fraud detection using machine learning algorithms— Study of customer behaviour. Grenze Int J Eng Technol (GIJET) 6(2):143–150 10. Khac NL, Markos S, Kechadi MT (2010) A data mining-based solution for detecting suspicious money laundering cases in an investment bank. In: Second international conference on advances in databases, knowledge, and data applications. IEEE Menuires, France 11. Lin-Tao L, Na J, Jiu-Long Z (2008) A RBF neural network model for anti-money laundering. In: International conference on wavelet analysis and pattern recognition. IEEE, Hongkong 12. Deng X et al (2009) Active learning through sequential design, with applications to detection of money laundering. J Am Stat Assoc 104(487):969–981 13. Salehi A, Ghazanfari M, Fathian M (2017) Data mining techniques for anti money laundering. Int J Appl Eng Res 12(20):10084–10094