Deep Learning in Finance

WebinarNovember 2021

Vishnu Kamalnath, Data Science Expert at McKinsey & Company, speaks with Dr. Hugo Bowne-Anderson, data scientist and educator at DataCamp and host of the podcast DataFramed. Hugo and Vishnu cut through the hype about data science, machine learning, and deep learning in finance by looking at practical applications and how they are helping the industry evolve.

Get an insider’s account of what has worked in finance, from machine learning models being used to classify fraudulent activity, to reinforcement learning being used for call center operations, to deep learning for credit writing and signature analysis. Whether you are interested in various applications of deep learning or the finance industry in particular, this webinar is for you.

You can find the slides here.

Summary

Deep learning is catalyzing change in the finance sector by allowing for enhanced data analysis and decision-making processes. Machine learning models are employed to detect fraudulent activity while reinforcement learning is used for improving call center operations, placing deep learning at the cutting edge of financial innovation. The webinar outlined specific applications such as using convolutional neural networks to spot forged signatures and implementing deep learning for credit underwriting. The dialogue stressed the significance of transparency in machine learning models, particularly in regulated sectors like finance. Effective data science in finance demands a mix of technical skills and sector knowledge, emphasizing the vital role of "interpreters" who connect data science and business requirements.

Key Takeaways:

Deep learning is reshaping fraud detection and credit underwriting in finance.
Convolutional neural networks can efficiently identify intricate patterns, such as falsified signatures.
Transparency is vital for machine learning models in regulated industries like finance.
"Interpreters" are indispensable for linking data science and business comprehension.
Reinforcement learning is proving useful in improving call center operations.

Detailed Analysis

Machine Learning Techniques for Fraud Detection

In the finance sector, deep learning has shown to be an effective tool for fraud detection. A significant example discussed was the use of convolutional neural networks by a major bank in Latin America to address a unique fraud problem involving falsified signatures on ...
Read More

pension loan applications. By examining genuine and forged signatures, the neural network identified key forgery indicators, such as the presence of ink blotches that appear when individuals trace their own signatures nervously. This automated detection process considerably improved the bank's ability to detect fraudulent activities. As Vishnu Kamalnath observed, "Deep learning drastically changed the way the bank operated," indicating the transformative effect of these technologies.

Deep Learning in Finance: Credit Underwriting Applications

Credit underwriting is another domain where deep learning is having an impact. In this context, multilayer perceptrons are employed to identify interactions between variables that were previously not captured by traditional models like logistic regression. By using deep learning as a feature discovery tool, financial institutions can improve their existing models, resulting in substantial performance enhancements without sacrificing transparency. As Vishnu Kamalnath explained, "The bank was able to close the gap pretty well without facing issues," by integrating insights from deep learning into their logistic regression models. This approach allows banks to utilize advanced analytics while complying with regulatory requirements.

The Role of Interpreters in Data Science

The webinar highlighted the significance of "interpreters" in the data science field, especially within finance. These individuals possess both business savvy and data science skills, enabling them to link technical teams and business stakeholders. Dr. Hugo Bowne-Anderson emphasized that "machine learning or AI in general can never be hands-off," highlighting the need for context-specific domain knowledge to effectively apply AI solutions. Interpreters play a vital role in ensuring that machine learning models are developed with a clear understanding of business objectives and constraints, facilitating successful implementation and adoption.

AI and Machine Learning in Call Center Operations

Reinforcement learning is emerging as a useful tool for improving call center operations in the financial industry. By analyzing customer-agent interactions, reinforcement learning algorithms can identify optimal staffing levels and improve customer-agent matching, enhancing overall efficiency and customer satisfaction. Vishnu Kamalnath noted the potential of reinforcement learning in this area, describing how it can "perform really well" in optimizing complex operational metrics. Although its application in traditional banking is still limited, reinforcement learning offers promising opportunities for enhancing service delivery and operational efficiency.