The rapid growth of financial data has paved the way for data science and machine learning to become an integral part of solving business challenges. It is pivotal in creating new revenue streams and streamlining operational costs.
As with all emerging fields, understanding and navigating the key challenges of data science will help businesses reap the returns from machine learning.
Each business case has a unique mix of statistical concepts. Familiarity with the building blocks of machine learning will give business a deeper appreciation for developing more robust machine learning strategies.
Lets take a look at the three building blogs of machine learning.
Supervised Learning
Supervised learning is a machine’s task of learning to predict an outcome, based on past data with known outcomes. The goal is for the machine to construct a formula based on the data set it has been trained to predict an outcome. Let’s take a real estate example of how you can predict the value of a three bedroom property by learning from past home sales data.
| Sqft Size | Bedrooms | District | Price |
| 2,100 | 5 | District 10 | S$2M |
| 1,500 | 4 | District 10 | S$1.5M |
| 1,500 | 4 | District 9 | S$1.8M |
| 1,000 | 3 | District 9 | S$1.4M |
| 1,200 | 3 | District 10 | ??? |
Here are some other business case studies in the financial industry where we can apply supervised learning:
| Other Financial Business Case Studies | Training Data |
| Will this customer default on a car loan? | Previous car loans paid or defaulted |
| How many customers apply for a credit card next quarter? | Previous months of credit card applications |
| How many customers will refinance their home loan in this month? | Previous months home loan redemption and refinancing |
Unsupervised Learning
Unsupervised learning is a machine’s task of looking for interesting patterns, based on past training data with no pre-defined outcomes. The goal is to perform discovery and detect underlying patterns in the data set. The table below provides an example of how unsupervised learning is a useful technique when there is no clearly defined target outcome.
| Date | Account | Class | Amount | Payment |
| 2/3 | Acct 1 | Retail | $150 | Cash |
| 4/3 | Acct 1 | Petrol | $80 | Credit Card |
| 14/3 | Acct 2 | F&B | $15 | Contactless |
| 14/3 | Acct 3 | Petrol | $5,000 | Credit Card |
| 14/3 | Acct 1 | F&B | $20 | Contactless |
| 16/3 | Acct 1 | Petrol | $80 | Credit Card |
There are several other examples of pattern recognition models commonly used in the banking industry:
- Cluster Analysis – data points that are similar and grouped together. For instance, finding customer behavior patterns for contactless spending in small F&B purchases. The table above gives an example of the basic fields for consideration.
- Anomaly Detection –unusual patterns in data. For example, card payments of large amounts that are out of the ordinary. This pattern modelling is commonly used in anti-money laundering and fraud detection.
- Association Discovery –consistent patterns in data. For example, Account 1 in the above table consistently has petrol payments of less than $100 that is purchased with a credit card.
Reinforcement Learning
Reinforcement learning is a machine’s task of navigating in a specific environment to maximize its performance. A machine or software agent adjusts its behavior based on feedback from the environment.
Organizations that focus on resource optimization, and managing operating cost, would find reinforcement learning strategies useful. For example, reinforcement learning may be useful in optimizing travel frequency of cash trucks for ATM top-ups, while constrained by manpower and truck availability.
Time series data from ATM cash withdrawals provide the machine with feedback to develop an optimized schedule for cash top-ups across the ATMs while maintaining a sustainable utilization rate of manpower and transport resources.
This article, written by Johnson Poh, was first published on Learn@IBF.
Facebook Comments