Back to Blog

How Banks can Leverage Deep Learning

As the user base and number of transactions per day keeps on increasing in the banking sector, top banking CIOs are constantly on the look out for new technologies that can help in running their banks better and smarter.

According to the federal trade commission of America, in 2018, the total number of fraud and identity thefts stood at 2.99 million and is expected to increase further in 2019.

Deep Learning in Banks

Deep learning is a subset of machine learning. Deep learning tries to mimic the human brain, processes data and creates models which are used to make inferences. Deep learning also has the ability to learn from unstructured or unlabeled data without any supervision (unsupervised learning).

Instagram: @Exit
Twitter: @ExitVisuals

Banks generate enormous amount of data on a daily basis. Most of this data is unstructured. It could take humans much more time to understand and extract relevant information from such data. With the help of machines, deep learning and the ability to scale at will using cloud computing, all this data can be processed much quicker.

Areas where banks can leverage deep learning

Non performing assets reduction

Bad loans (or NPAs) are loans that are not paid by the borrower. Deep learning algorithms can predict accurately the credit worthiness of a customer by leveraging multiple data sources thereby mitigating risk. They have also created the ability to use customer data to understand the behavioral economics of each customer and provide personalized collection and recovery process.

Related: Download the checklist that lays out everything you should consider  before implementing a machine learning project or workflow →

Fraud prevention

In 2018, the Federal Trade Commission processed 1.4 million fraud reports totaling $1.48 billion in losses. According to the FTC’s “Consumer Sentinel Network Data Book”, the most common categories for fraud complaints were impostor scams, debt collection and identity theft. Credit card fraud was most prevalent in identity theft cases - more than 167,000 people reported a fraudulent credit card account was opened with their information. With deep learning such incidents can be more effectively prevented.

Let's take for example, a bank trying to combat money laundering. A traditional approach might work on rules such as the transaction amount, past history, location of the transaction. Even traditional machine learning approaches may falter here due to the lack of structured or labeled data.

Deep learning excels at identifying patterns in unstructured data (text, images, sound or video). A deep learning technique might take in factors such as:

  1. Time

  2. Location

  3. IP address

  4. Device type

  5. Last transactions

And more factors to classify a given transaction as a fraudulent one or not in real time. When a given transaction is classified as fraudulent, the account of the person in question could be frozen until further action.

Personalization

With the advent of the internet and mobile phones, the quantum of customer interactions with banks has increased multi fold over channels such as website, call centers, ATMs and kiosks. More data makes that the prediction models built by the deep learning algorithms more accurate at segmenting customers into groups which can be used to personalizing their banking experience such as product recommendations and customer service. This can have a direct impact on profits.

Visit Skyl.ai to learn more about different deep learning and computer vision projects.

how-to-ensure-that-your-machine-learning-project-is-successful

    

Comments