**2. Emerging application of machine learning in finance**

With the increasing availability and declining cost for complex models executing on high-power computing devices exploiting the unlimited capacity of data storage, the financial industry is geared up to exploit the benefits of machine learning to leverage a competitive business edge. While some of the use cases have already found their applications in the real world, others will need to overcome some existing business and operational challenges before they are deployed. Some of the applications are mentioned below.

**Risk Modeling:** One of the major applications of AI/ML models and algorithms is in the extensive domain of risk modeling and management [1]. While on one hand, the risk modeling credit and market is a critical application of machine learning, on the other hand, a non-finance application such as operational risk management, compliance, and fraud management is also quite important. The majority of the classification approaches and modeling techniques in machine learning such as binary logistic regression, multinomial logistic regression, linear and quadratic discriminant analysis, and decision trees, etc., are the foundational building blocks of applied modeling in the real world. However, in data science applications, the availability of data and its richness play a pivotal role. Hence, in data-rich applications such as credit risk modeling and scoring, designing mortgage schemes, the AI/ML models have already made substantial inroads in comparison to scenarios such as low default credit portfolios for well-known parties that lack the availability of data. Fraud analytics remains another intensive area of AI/ML applications in the non-financial domain.

**Portfolio Management:** The portfolios are designed based on the recommendations of smart algorithms that optimize various parameters with return and risk being the two most important ones [2]. Using the information provided by the users such as their ages of retirement, amount of investment, etc., and other associate details such as their current ages, current assets at hand, the algorithm allocate the invested amount into diverse asset classes to optimize the return and the risk associated with the portfolio. Once an initial allocation is made, the algorithm continuously monitors the market environment and changes the allocation so that the portfolio remains at its optimized level always. These AI-enabled portfolio managers, known as the Robo-advisors are increasingly being used in real-world portfolio design due to their superior adaptability and optimization skill to their human counterparts.

**Algorithmic Trading:** Algorithmic trading exploits the use of algorithms to carry out stock trading in an autonomous manner with minimal human intervention [3]. Invented in 1970, algorithmic trading deploys automated pre-programmed stock trading instructions that can be executed very fast and at a very short time interval (i.e., at a very high frequency) to optimize trading returns and objectives. Machine learning and artificial intelligence have pushed algorithmic trading into a new dimension where not only advanced trading strategies can be made very fast but also deep

### *Introductory Chapter: Machine Learning in Finance-Emerging Trends and Challenges DOI: http://dx.doi.org/10.5772/intechopen.101120*

insights can be made into the stock price and overall market movements. While most hedge funds and financial organizations do not make their trading strategies public, it is well known that machine learning, of late, is playing an increasingly important role in calibrating high-frequency, high-volume trading decisions in real-time for critical applications.

**Fraud detection and analysis:** Fraud detection and analysis is one of the most critical machine learning applications in the finance industry [4]. There is an increased level of security and privacy risk associated with sensitive information both from the organization and personal front due to ubiquitous availability of connectivity, high computational power in devices, an increased amount of data stored and communicated online. These issues have changed the way online fraud analysis and detection are being made. While detection in earlier days used to depend on matching a large set of complex rules, the recent approaches are largely based on the execution of learning algorithms that adapts to new security threats making the detection process more robust while being agile.

**Loan and Insurance Underwritings:** Loans, credit, and insurance underwriting is also an area where the large-scale deployment of machine learning models can be made by financial institutions for achieving competitive advantage [5]. At large banks and insurance firms, the availability of historical data of consumers, financial lending/borrowing information, insurance outcomes, and default-related information in paying debts, can be leveraged in training robust machine learning models. The learned patterns and trends can be exploited by the learning algorithms for lending and underwriting risks in the future to minimize future defaults. The deployment of such models can lead to a paradigm shift in business efficiency and profit. However, at present, there is a limited utilization of such models in the industry as their deployments are largely confined within large financial institutions.

**Financial Chatbots:** Automation in the finance industry is also an outcome of the deployment of machine learning and artificial intelligence. Accessing the relevant data, machine learning models can yield an insightful analysis of the underlying patterns inside them that helps in making effective decisions in the future. In many cases, these models may provide recommended actions for the future so that the business decision can be made in the most efficient and optimum way. AI-based systems in financial applications also can minimize their errors learning fast from their past actions that also reduce wastages of precious resources including time. AI chatbots provide an effective way of interaction with the customers while automating many routine tasks in a financial institution [6].

**Risk Management:** Machine learning techniques are revolutionizing the way corporates handle the risks associated with their operations. Risk management examples are diverse and infinite ranging from deciding about the amount a bank should lend a customer or how to improve compliance to a process or the way risk associated with a model can be minimized [7].

**Asset price prediction:** Predicting future asset prices is one of the most challenging tasks that finance professionals have to do frequently. The price of an asset is affected by numerous factors driven by the market including speculative activities. In the classical approach, an asset price is determined by analyzing historical financial reports and past market performances. With rich data available, of late, machine learning-based models have started playing significant roles in predicting future asset prices in a robust and precise manner [8].

**Derivative pricing:** In the traditional approach, derivative pricing models are designed on numerous assumptions which do not hold good in the real world. These assumptions attempt to establish an empirical relationship among several predictor variables such as the strike price, maturity time, and option type, and the target variable which is the price of the derivative. Machine learning models have gotten rid of the requirement of such assumptions as they attempt to fit in the best possible function between the predictors and the target by minimizing the error. Accuracy and the minimal time needed for the deployment of the models in real-world use cases make machine learning the most impactful paradigm in the task of derivative pricing [9].

**Money Laundering:** As per a report published by the United Nations, it is estimated that 2–5% of the world's aggregated GDP is accounted for the amount of money being laundered worldwide. Machine learning models can find applications in detecting money laundering activities with minimal false-positive cases [10].
