APT 3025: APPLIED MACHINE LEARNING
INSTRUCTOR: PROF. JOHN KIMANI
GROUP ASSIGNMENT 4
FINTECH PRESENTATION INSIGHTS
MEMBERS
MILDRED MARGARET- 665455
ANGELA NDUNG’U - 666663
TEDDLA PENINAH - 666624
JANE ODAWA 666077
AMINA SORA - 665774
KYLA MASOUD - 666900
1) Executive Summary...................................................................................................................2
2) Key Insights from the Presentations........................................................................................ 3
2.1 AI and Machine Learning in Fraud Detection and Financial Analytics.............................. 3
2.2 The Future of Decentralized Finance (DeFi) Using Blockchain in Bitcoin........................ 4
a) Transparency and Security: The immutable ledger............................................................... 4
b) Smart Contracts: Automating Finance without Intermediaries............................................. 4
c) Bitcoin as an Alternative Financial System:..........................................................................5
3) Machine Learning in Fintech Applications.............................................................................6
a) Fraud Detection......................................................................................................................6
b) Personalized Banking............................................................................................................ 6
c) Credit Scoring & Risk Assessment........................................................................................7
d) Algorithmic Trading.............................................................................................................. 8
e) Emerging Applications.......................................................................................................... 8
4) Challenges and Ethical Considerations in Fintech AI........................................................... 9
4.1 Key Challenges in Implementing ML in Fintech................................................................ 9
a. Data Privacy & Security...................................................................................................9
b. Biases............................................................................................................................. 10
c. Regulatory Uncertainty Challenge:................................................................................10
4.2 Ethical Considerations....................................................................................................... 11
a) Transparency:.................................................................................................................11
b) Accountability:.............................................................................................................. 11
c) Inclusivity:..................................................................................................................... 11
5) Conclusion and Personal Reflection...................................................................................... 12
Conclusion............................................................................................................................... 12
Key Points with an Intersection with My Studies:...................................................................12
Significance Relevant to My Career Goals..............................................................................12
1
Fintech Presentation Insights: The Role of Machine Learning in Financial Technology
1) Executive Summary
On 6th March 2025, Dr. Ana Armstrong of Greenwich University was the keynote speaker at
USIU-Africa, with the address "The Role of Technology in Finance", at the School of
Humanities and Social Sciences, USIU-Africa. It highlighted how modern technologies,
especially machine learning (ML), artificial intelligence (AI), and blockchain, have affected
business cycles within the financial sector.
These were the three main pillars of Dr. Armstrong's address:
● Democratizing Finance - The impact of technology on access to financial services.
● Bitcoin and Blockchain - How decentralizing technologies are influencing traditional
finance.
● Emerging Fintech Trends - The role of AI, machine learning, and digital currencies in the
future of finance.
This occasion was also the platform for a panel discussion in which other participants, industry
experts, and academics shared their thoughts on the regulatory challenges, ethical issues, and
practical application of fintech innovations. This report brings together various key aspects
discussed in the presentations, correlates some of them with the course concepts of machine
learning, and finally discusses implications for the future of financial technology.
2) Key Insights from the Presentations
2.1 AI and Machine Learning in Fraud Detection and Financial Analytics
Among the insights of Dr. Armstrong, the increasing reliance upon machine learning in fraud
detection and assessment of financial risks stands out. She explained how traditional rule-based
systems for fraud detection have been replaced by supervised and unsupervised models that
apply machine learning as they scale up to real-time processing of large data sets.
a) Supervised Learning for Fraud Classification
2
Algorithms like Random Forests, Gradient Boosting Machines (GBM), Neural Networks, etc.,
are all supervised in that they learn on labeled datasets that pertain to the classification in
question (i.e., fraudulent vs. non-fraudulent transactions).
For instance, banks would apply such models to flag potentially fraudulent transactions-a case in
point would be an immediate large-value transfer-looking for patterns of prior fraud. A number
of variables may be considered for assessment.
b) Unsupervised Learning for Anomaly Detection
Isolation Forests and Autoencoders among others, are used for detecting unusual behavior
without any predefined label.
For instance, PayPal could be preventing account takeovers with anomalies such as logins from
unknown locations.
2.2 The Future of Decentralized Finance (DeFi) Using Blockchain in Bitcoin
Dr. Armstrong also elaborated on the scope of blockchain technology in restructuring the
financial systems. She stated;
a) Transparency and Security: The immutable ledger
The immutable ledger Distributed Ledger Technology (DLT) or blockchain records all
transactions over the computers spread in a decentralized mode. Such benefits were mentioned
by her.
i) Tamper-Proof Records:
Once recorded in the blockchain, the transaction cannot be interfered with or deleted, thanks to
consensus mechanisms such as Proof of Work or Proof of Stake and the process of cryptographic
hashing. For instance, Walmart uses blockchain technology for tracking their food shipments in
the supply chain finance, as this assists with authentication and fraud prevention.
ii) Auditability:
3
The openness of every transaction to the public reduces the need for third-party audits.
For instance, with Bitcoin, anyone can track the flow of transactions (e.g., tracing donations to
Ukraine during the war).
b) Smart Contracts: Automating Finance without Intermediaries
According to Dr. Armstrong, the use of smart contracts-self-executing agreements coded on
blockchains like Ethereum-can prove to be a very important factor in revolutionizing financial
services.
How Smart Contracts work:
When conditions that are predefined are met, smart contracts carry out their operations. They
execute the payment of insurance automatically when the flight is confirmed to be delayed.
For example: Lending to instant consumers without banks - These smart contracts are deployed
together into an end-to-end model by Aave (a DeFi platform).
c) Bitcoin as an Alternative Financial System:
For example, Dr. Armstrong also said into that specific section concerning how the use of
BitCoin in hyperinflationary economies and poor currency environments shown it as:
i) A Store of Value:
Citizens of countries like Venezuela and Zimbabwe see BitCoin as one of the ways to save their
income during currency collapse.
Example: Demand for Bitcoin in Argentina rose once the peso had depreciated by 90% against
the dollar over ten years.
ii) Cross-Border Payments:
Cheap remittances for migrant workers in turn (like the case of Bitcoin in El Salvador).
Use Cases in Decentralized Finance (DeFi):
● Applications in Decentralized Finance (DeFi):
4
● Lending/Borrowing-The Compound Finance enables a user to generate interest from
deposits using smart contracts.
● Insurance- Etherisc provides automated crop insurance for farmers.
● Trading- Uniswap allows users to swap tokens directly, trustlessly and without a central
authority.
3) Machine Learning in Fintech Applications
Machine learning (ML) applications for financial services, according to Armstrong, promise
speed, accuracy, and personal touch. The following is an in-depth discussion of some key areas
of application that she addressed, including technical details, case studies, and business effects.
a) Fraud Detection
Machine Learning Techniques Used:
● Random forests and gradient boosting methods (XGBoost): ensemble techniques
incorporating multiple decision trees and aimed at the detection of anomalous
transactions.
● Neural networks (deep learning): for pattern recognition in high-dimensional data, such
as history of transactions.
● Anomaly detection (isolation forests, autoencoders): detect outliers in real-time payment
flows.
Real-World Example: Visa AI Fraud Scoring System
With ~200bn transactions processed each year, Visa uses machine learning to:
● Assign a fraud risk score (from 0 to 99) to fraudulent transactions within a couple of
milliseconds.
● Reduce false declines (legitimate transactions flagged as fraud) by 30%, saving
merchants $2.4bn per year.
Business Impact:
Cost Savings: Reduction of fraud-related losses (credit card chargeback).
5
Customer Assurance: Better user experience with respect to false alarms.
b) Personalized Banking
Machine Learning Techniques Used:
● Clustering (K-means): create segments of customers based on spending habits.
● Recommender Systems (collaborative filtering): recommends financial products for them
(loans, credit cards).
● NLP: powers AI chatbots (Bank of America’s Erica) in answering customer queries.
Real-World Example: AI-Powered Wealth Management
● Betterment & Wealthfront leverage ML to:
● Automate portfolio rebalancing with changing market conditions.
● Offer tax-loss harvesting strategies.
Business Impact:
● Increased Engagement: Personal recommendations will help with customer retention.
● Operational Efficiency: Reduce dependency on human financial advisors for mundane
tasks.
c) Credit Scoring & Risk Assessment
Machine Learning Techniques Used:
● XGBoost/LightGBM: Finds out non-linear relationships in credit data (payment history,
social media activity, etc.).
● Logistic Regression: Assessing probability of loan applicants default.
● Graph Neural Networks (GNN): Analyze linkages amongst borrowers (detecting loan
fraud rings).
Application in Real Life: Alternative Credit Scoring
Tala (Kenya) and Upstart (USA) have machine-learning systems to score:
6
● Thin-file borrowers (no credit history) using mobile phone usage and utility bill
payments.
● Approve 40% more loans without increasing default rates.
Business Impact:
● Financial Inclusion: Increasing access to credit for underserved populations.
● Risk Mitigation: More accurate than traditional FICO scores.
d) Algorithmic Trading
Machine Learning Techniques Used:
● Reinforcement Learning-the AI agent has to learn how to optimize trading strategies
through (e.g., stock picks through Q-learning).
● Time-Series Forecasting-(LSTM(s), Prophet): forecasting price movements based on
historical market data.
● Sentiment Analysis-(NLP): Sentiment indicators derived from news and social media.
Real-World Examples: Hedge Fund Strategies
Renaissance Technologies is using ML for:
● HFT executed in microseconds.
● Returns of about 35 percent per annum (about 10 percent for the S&P 500).
Business impact:
● Profitability: ML clearly beats human traders in panic situations.
● Liquidity: To a slight extent, liquidity improves by arbitraging via automation.
e) Emerging Applications
The following emerging use cases were also mentioned by Dr. Armstrong:
● Generative AI for Synthetic Data: Producing a realistic financial data set for stress testing
of models.
7
● Explainable AI: Assists regulators in auditing black-box ML models (e.g., SHAP values
for loan approvals).
● Quantum ML: Researched by JPMorgan for portfolio optimization (still in the
experimental form).
4) Challenges and Ethical Considerations in Fintech AI
4.1 Key Challenges in Implementing ML in Fintech
With the opportunity of transforming machine learning (ML) into another discipline, an enduring
challenge to FinTech is posed at having to accommodate secure, fair, and compliant systems.
Below are some of the embraced challenges with areas of concern, some illustrations from actual
settings, and possible solutions.
a. Data Privacy & Security
Financial institutions have to deal with sensitive information such as transaction records, credit
ratings, and personal identification data in bulk. A possible case of invasion in data privacy can
lead to crimes like identity theft and financial fraud, rendering reputations of the offending
institution bad.
Real Life Example: Equifax data breach, 2017: The Equifax data breach harmed the personal
information for 147 million customers and comprised Social Security numbers and even more
credit card numbers.
Solutions:
● Federated Learning: Train ML models without the exchange of raw data on multiple
devices/servers. For example, banks may work together to develop or improve their
respective fraud detection models while keeping customer data private. For example,
Gboard by Google.
● Homomorphic Encryption: To preserve privacy through computation on encrypted data,
this involves protection of data while allowing some processing on it or using it. For
8
example, JPMorgan's Cryptonets applies homomorphic encryption to credit scoring,
enabling computations on encrypted financial data.
b. Biases
Biases can enter and amplify in machine learning models created using historical data, leading to
discriminatory outcomes in loan approval, hiring, and underwriting for insurance.
Real-World Examples:
● Apple Card Gender Bias (2019): The algorithm applied by Apple Card treated men and
women with equal credit profiles differently: men much favored when it comes to
granting higher credit limits. This bias hence led to criticism.
● Amazon AI Recruiting Tool: Trained on hundreds of thousands of resumes submitted
over the years, most of which came from men, the tool models past hiring patterns that
discriminate against women in its recommendations. The discrimination toward women
here was enhanced by the historic underrepresentation of women in technology.
Mitigation Strategies
● Fairness-Aware Algorithm - To pinpoint accuracy in identifying or rectifying bias on both
the data and the model sides.
● Diverse Dataset Curation - To ensure that the training data captures underrepresented
groups-like low-income borrowers would be necessary.
c. Regulatory Uncertainty Challenge:
Evolution is rapid; new technology is emerging every day within the fintech space, most recently
with DeFi and CBDCs; regulatory frameworks cannot keep pace, and this has caused legal
ambiguities.
Case Studies:
9
SEC v. Ripple (2023): The Securities and Exchange Commission sued Ripple Labs for offering
XRP tokens as unregistered securities. This is so because the existing regulation has no clear
definition on cryptocurrencies.
Most DeFi protocols are also unlicensed under traditional financial laws, raising questions on
their consumer protection and anti-money laundering status.
4.2 Ethical Considerations
Fintech must address ethical challenges through;
a) Transparency:
Black-box AI methods (e.g., deep neural networks) are inherently unexplainable, and as such,
explaining business decisions taken before customers or regulators becomes quite problematic.
Solution:
Use explainable artificial intelligence (XAI) techniques to explain model outputs.
b) Accountability:
Finding particular instances of wrong decisions becomes extremely challenging in situations of
mistakes made by the AI system (for instance, wrongful loan denials).
Framework: An established system of audit trails and human intervention or review must be
created in very serious instances.
c) Inclusivity:
Fintech must make sure that the AI services do not sideline any marginalized groups.
For instance:
Alternative credit models that use nontraditional scoring data able to capture this information
(for instance, utility payments) do a lot of good in granting access to financial services.
10
❖ Dr. Armstrong’s Take: “Ethics must be embedded in fintech development—not
treated as an afterthought.”
5) Conclusion and Personal Reflection
Conclusion
Dr. Armstrong's keynote speaking engagedly highlighted how machine learning and blockchain
can genuinely change the face of finance and legitimately illuminate pathways intersecting with
my studies in Computer Science, especially AI/ML and Cybersecurity.
Key Points with an Intersection with My Studies:
The discussion on supervised and unsupervised learning, for instance, fraud detection using
Random Forests versus anomaly detection using autoencoders, directly reinforced some content
from my AI-Machine Learning class.
My curiosity had been about the relationship between algorithmic fairness and ethical AI, with
special emphasis on bias mitigation and fairness-aware algorithms, as in IBM's AI Fairness 360.
Cybersecurity Connections: From this standpoint, privacy-preserving ML, such as federated
learning or homomorphic encryption, connects back with my focus on Cybersecurity by
demonstrating some sort of assurance in guaranteeing security with sensitive data where AI is
involved in a subset of fintech.
The coding aspect taught us about the coding flaws try to mitigate in Poly Network's case for
instance, the architecting of a highly immutable traceable ledger with a smart contract payment
system.
11
Significance Relevant to My Career Goals
As a Computer Science student specializing in Cybersecurity, I would view fintech as a crucial
area where my skills can be of use for:
● Secure AI Deployments: Secure AI and Machine Learning models from adversarial
attacks in finance, such as data poisoning types.
● Privacy-Preserving Solutions: The identification of such transactions is blocked using
techniques like zero-knowledge proofs (ZKPs).
● Regulatory Tech (RegTech): Automated tools (e.g., productive AML checks) enable
insight into the operation.
Dr. Armstrong's lecture gave me the motivation to further explore the world of FinTech and
hopefully contribute my skills in cybersecurity toward solving some of the major challenges
facing the FinTech area.
12
