Search

Artificial Intelligence & Machine Learning

Machine Learning for Fraud Fighting:
Make it real time and prescriptive

The solution relies on the use of deep learning algorithms and of the probability of fraud to block suspicious loans.

Enhancing Antifraud Experience

WHICH IS THE WORST PART OF BEING A COMPANY LENDING MONEY TO PEOPLE?

A never ending challenge with more and more “professional” defrauders who keep changing their habits, breaking the rules, reinventing the game to keep earning from their crimes.

To identify and anticipate fraud detection we are using advanced technologies, functional, statistical and technical deep know how in order to implement an efficient Machine Learning algorithm. Pattern Recognition, statistical modelling, neural networks are few keywords that describe what we have done. Deep Learning, prescriptive analytics, real time online scoring are few words that describe what we are designing with the customer.

Target Reply supported one of the first Consumer Credit Company in Italy with millions of loans every year. The customer delivers credit through different product lines: retail loans, personal loans, credit cards and leasing.

GOALS

Main customer’s goals were:

  • Reduce loss money from frauds
  • Block sospicious loans
  • Quickly create predictive rules for fraud detection

Target Reply‘s solution anticipates and automates fraud detection: by means of deep learning algorithms identifies serial fraudsters that change their habits to evade controls and creates more advanced and predictive models that fit in new and unknown contexts.

FOCUS ON PREDICTION & PRESCRIPTION

Target Reply’s approach is to develop Machine Learning models such as neural networks, logistic regression or decision tree, compare results and choose the best model to promptly identify potential fraudsters.
The solution also has a real time prescription component able to run the machine learning algorithms in the loan acceptance phase, so making the overall solution even more effective. The development of new components will rely on the use of deep learning algorithms for image recognition thus involving unstructured data into the analysis.