report

PAYPAL MACHINE LEARNING

JOAO ALEMANY AND KARIM BARAKAT

CONTENT 01

02

03

04

INTRODUCTION

APPLICATIONS IN PAYPAL AND OTHER PLAYERS

CHALLENGES

HOW IT CAN BE IMPROVED

Introduction PayPal is a company founded in the USA in December of 1998. It was founded by Max Levchin, Peter Thiel, and Luke Nosek. It was originally named Fieldlink, and was later renamed as Confinity. It originally was company that developed a security software for handheld devices. However after doing terribly for a year it switched to what it is now known for which is a digital wallet. In March 2000, they merged with x.com a financial service company founded by Elon Musk and his partners. Later that year Musk decided to focus only on payments. Musk was later replaced by Peter Theil as CEO of x.com which he then renamed to PayPal. In 2002 PayPal went Public and had an IPO ticker listed under PYPL. Shortly later, PayPal was bought by eBay for 1.5 Billion dollars in eBay stock. Forwarding 13 years in 2015, eBay spun off PayPal to its shareholders and it became and independent company again.

Machine Learning What is Machine learning? Machine learning is a field within the AI industry. It uses Algorithms to provide computers with the ability to use mass data in order to identify patterns in order and to come up with predictive analysis. This method allows computers to preform tasks alone and without the obligation to be programmed. The term was first used in 1959, however it became more popular recently as the capabilities of computers increased and there is more available data. Machine Learning is divided into 3 categories: Supervised Learning: Has prior leaning incorporated in them, they are based on a tag system which collaborates with data that allows them to make predictions. Unsupervised Learning: Does not have prior knowledge, they face a data chaos with the goal of finding patterns, that will somehow allow the machine to make predictions. Reinforced Learning: Its objective is for an algorithm to learn from its own experience, and correct itself. it is not as common as the other two though.

Fraud Detection

MAIN APPLICATIONS AT PAYPAL

Risk Management

Customer Experience

FRAUD DETECTION

PayPal gathers vast amounts of data on every transaction, including:

User information (name, location, IP address) Transaction detai ls (amount) Device information Past transaction history of the user and simi lar users

Data collection and analysis

This data is used to train models to identify patterns associated with fraudulent activity , such as:

Unusual spending habits ( large purchases compared to past behavior) Mult iple accounts l inked to the same device

Training ML Models

I f a unl ikely behavior or any suspicious activity is detected, the model wi l l g ive a fraud score to the transaction.

Depending on the score, there might have a review on the transaction and some actions can be made, as requesting addit ional information or suspending the transaction

Fraud Detection

FRAUD DETECTION

CREDIT RISK MANAGEMENT

Again, it ’s created a Machine Learning model uti l iz ing data from the users to detect patterns.

This data include: f inancial and demographic information

With this data the users are given scores, that determine the probabi l ity of default .

Model training

This Model is able to improve decision making and give more accurate analysis .

Also, with this model it is possible to improve eff iciency by reducing the t ime to make those analysis and saving resources.

Furthermore, the model helps to reduce human decision making biases.

Benefits

Automated customer service: Chatbots Faster transactions : optimize transaction processing

identifying and priorit iz ing high-risk transactions for manual review, al lowing low-risk transactions to be processed more quickly .

Improved search and navigation

CUSTOMER EXPERIENCE

Recommendations: ML can analyze customer data (purchase history, browsing behavior) to recommend relevant products and services, l ike f inancial products Dynamic content: ML can personal ize website and app content , showcasing information and features relevant to each user 's profi le and behavior .

Personalization Efficiency

CHALLENGES

01

ETHICS AND BIAS

02 03

TRANSPARENCY DATA SECURITY AND PRIVACY

CHALLENGE Data: Depending on the data used to train the model , i t might create biases in the model 's outputs. For example, a model trained on data from individuals with high credit history might disadvantage users with l imited credit history .

Algorithmic bias: Even with balanced data, biases can be made during the training process, such as denying certain users access to credit or other f inancial services based on factors l ike race, gender , or location.

SOLUTION Data collection strategies: seek data from underrepresented groups and demographics to ensure a more balanced data base for training ML models

Review process: integrate human review processes when the decision making has a high impact for the individuals

Fairness-aware algorithms : a lgorithms designed to consider fairness metrics during training, mit igating bias in model outputs

ETHICS AND BIAS

CHALLENGE Machine Learning models are very complex and diff icult to understand and explain it ’s decision making

This lack of transparency may raise concerns about it ’s decision making, for example, why the model would deny credit to a cl ient and if it has any bias or not.

This could turn into a regulatory violation and cause problems to the company and customers.

SOLUTION

TRANSPARENCY

There isn’t a way to be sure about the decision making, due to it ’s complexity , but that are some ML models that tr ies to helps with that .

Explainable AI : Uses some techniques to help users understand factors inf luencing model decisions and identify potential biases, promoting transparency and trust .

CHALLENGE Another concern is about cybersecurity , s ince these information used by the Machine Learning models are valuable, there’s a high r isk of attack attempts.

A data leak could bring great f inancial and reputational impact to the company and the loss of the consumer confidence, due to regulatory violations

SOLUTION

DATA SECURITY AND PRIVACY

Investments in cybersecurity

Penetration testing: involves hir ing a hacker to attempt to steal the company's information in order to identify system vulnerabi l it ies. This process helps to understand system weaknesses, correct them, and enhance system security.

Regular security auditions