Figure 1 illustrates the transaction graph generated by the system. Nodes represent entities and transactions, while connections represent relationships between them. Red triangle nodes represent transactions classified by the model as fraudulent or high-risk.
Figure 1. Example visualization of the graph-based fraud detection network.
Machine Learning · Telecom Fraud · Cybersecurity · Intelligent AI Systems
Independent AI Research & Engineering
I build and experiment with practical artificial intelligence systems for real-world problems, with a particular focus on:
- Telecom fraud detection
- Cybersecurity analytics
- Anomaly detection
- Machine-learning benchmarking
- Explainable AI
- Autonomous AI systems
- Fraud and risk intelligence
My work combines 20+ years of telecommunications revenue assurance and fraud-management experience with modern machine-learning and AI engineering.
The central question behind much of my work is:
How can intelligent systems detect complex, evolving patterns that traditional rules struggle to identify?
ARTIFICIAL INTELLIGENCE
│
┌───────────────────┼───────────────────┐
│ │ │
▼ ▼ ▼
MACHINE LEARNING ANOMALY DETECTION AI SYSTEMS
│ │ │
└──────────────┬────┴────┬──────────────┘
│ │
▼ ▼
TELECOM FRAUD CYBERSECURITY
│ │
└────┬────┘
▼
INTELLIGENT RISK
ANALYTICS
Machine Learning
- Supervised learning
- Ensemble learning
- XGBoost
- Random Forest
- Classification
- Model benchmarking
Anomaly Detection
- Isolation Forest
- Autoencoders
- Behavioral modelling
- Novel fraud detection
- Unlabelled-data analysis
Telecom AI
- IRSF
- Wangiri
- SIM-box / bypass fraud
- Subscription fraud
- Account takeover
- A2P / SMS fraud
- Revenue assurance
Cybersecurity AI
- Cyber-fraud detection
- Behavioral threat detection
- Ransomware detection
- Security analytics
- Risk scoring
My approach is not limited to using machine-learning libraries. I am interested in the underlying mathematical and statistical mechanisms that allow models to distinguish normal from fraudulent behavior.
[ P(y=1|x)=\sigma(\beta_0+\beta^Tx) ]
where:
- (x) = feature vector
- (\beta) = learned parameters
- (y=1) = fraud
- (\sigma) = sigmoid function
[ Precision=\frac{TP}{TP+FP} ]
[ Recall=\frac{TP}{TP+FN} ]
[ F_1= 2\frac{Precision\times Recall} {Precision+Recall} ]
[ E(x)=||x-\hat{x}||^2 ]
A sufficiently large reconstruction error can indicate behavior that differs substantially from the patterns learned from legitimate activity.
[ s(x,n)= 2^{-\frac{E(h(x))}{c(n)}} ]
where shorter isolation paths indicate observations that are easier to separate from the normal population.
My projects generally follow a repeatable research pipeline:
PROBLEM
│
▼
DATA COLLECTION
│
▼
DATA EXPLORATION
│
▼
FEATURE ENGINEERING
│
▼
MODEL DESIGN
│
▼
EXPERIMENTATION
│
▼
MODEL BENCHMARKING
│
▼
EXPLAINABILITY
│
▼
DEPLOYMENT
│
▼
OBSERVATION & ITERATION
The objective is to move beyond:
"I built an AI application."
toward:
"I investigated a problem, tested multiple approaches, measured their behavior and converted the findings into a working system."
A security monitoring application designed to detect suspicious or unauthorized activity and provide timely security alerts through a practical and intuitive interface.
|
|
|
Application: Intrusion Detection & Security Alerts
Status: Developed
An interactive cybersecurity and fraud-awareness platform developed by AI Labs, using scenario-based challenges, scoring, gamification, progress tracking, achievement badges, and certification.
Key capabilities:
- 🎣 Phishing awareness
- 📱 SMS fraud detection
- 🧠 Social engineering scenarios
- 🔐 Ransomware response
- ☎️ Telecom fraud awareness
- 🤖 AI threat challenges
- 📊 Cyber awareness scoring
- 🏆 Achievement badges
- 🎓 Certification
🌐 Live Demo:
https://adnanmalik83.github.io/CyberAware-AI/
💻 Repository:
https://github.com/adnanmalik83/CyberAware-AI/
A machine-learning benchmark investigating different approaches to telecom fraud detection.
Algorithms explored:
- Logistic Regression
- Decision Trees
- Random Forest
- XGBoost
- Isolation Forest
- Autoencoders
Which machine-learning technique is most appropriate for different telecom fraud scenarios?
The benchmark compares supervised and unsupervised approaches rather than assuming that one algorithm is universally optimal.
Focus:
Classification · Anomaly Detection · Model Comparison · Fraud Analytics
A machine-learning cybersecurity analytics platform designed to identify fraudulent and high-risk activity.
Technology:
- XGBoost
- Random Forest
- Feature engineering
- Feature importance
- Risk classification
- Streamlit analytics
Research direction:
Applying machine learning to behavioral cybersecurity signals and fraud-event classification.
An AI-driven platform for detecting and analysing fraudulent A2P SMS activity.
The system explores:
- XGBoost
- Random Forest
- Fraud classification
- Operational analytics
- Risk scoring
- Explainable model outputs
The project is particularly focused on translating machine-learning predictions into an operational fraud-management workflow.
An open AI-oriented exploration of telecommunications fraud detection and intelligent fraud analytics.
Research domain:
Telecom AI · Fraud Detection · Revenue Assurance · Machine Learning
An experimental AI research pipeline exploring local-first research workflows and automated information analysis.
Research themes:
- AI research workflows
- Automated information retrieval
- Local-first processing
- Research automation
An AI-driven data-analysis system exploring how analytical workflows can be automated.
The project investigates the transition from:
Human → Query → Analysis → Visualization
toward:
Human
↓
AI Analyst
↓
Data Understanding
↓
Analysis
↓
Visualization
↓
Insights
An AI-oriented cybersecurity project focused on mobile security and threat awareness.
Technology direction:
Android · Kotlin · Cybersecurity · AI
A cybersecurity application designed to provide users with practical security monitoring, threat awareness, and protection capabilities through an intuitive interface.
.jpg&engine=chrome)
|
.jpg&engine=chrome)
|
.jpg&engine=chrome)
|
Application: Cybersecurity & Threat Protection
Status: Developed
An experimental cybersecurity project investigating ransomware detection through canary mechanisms and log analysis.
Research direction:
Canary Signals
+
Log Behaviour
+
Anomaly Detection
↓
Early Ransomware Detection
| Project | Research / Engineering Area |
|---|---|
| Telecom Fraud AI Benchmark | ML benchmarking |
| AI Cyber-Fraud Detection | Cybersecurity AI |
| A2P SMS Fraud Operations Center | Telecom fraud |
| OpenTelecomFraudAI | Telecom AI |
| Local-First AI Researcher | AI research automation |
| Autonomous AI Data Analyst | Autonomous analytics |
| CyberShield | Mobile cybersecurity |
| Ransomware Canary | Ransomware detection |
| SmartRecruit-AI | NLP / semantic matching |
| AI SEO Optimizer | Applied AI automation |
One of my core research interests is understanding that different fraud mechanisms require different analytical approaches.
| Fraud Type | Potentially Suitable Approach |
|---|---|
| IRSF | XGBoost / Random Forest |
| Wangiri | XGBoost + Rules |
| SIM-box / Bypass | Random Forest + Anomaly Detection |
| Subscription Fraud | XGBoost / Random Forest |
| Account Takeover | XGBoost + Behavioral Anomaly Detection |
| A2P / SMS Fraud | XGBoost / Random Forest |
| Unknown Fraud | Isolation Forest / Autoencoder |
| Limited Labels | Anomaly Detection |
| Highly Explainable Decisions | Logistic Regression / Decision Tree |
The goal is not to identify a universally "best" algorithm.
The goal is to identify the most appropriate analytical technique for the characteristics of the fraud problem.
Traditional fraud management often follows:
EVENT
↓
RULE
↓
ALERT
↓
INVESTIGATION
My research direction explores:
TELECOM EVENTS
│
▼
FEATURE ENGINEERING
│
┌─────────────────┼─────────────────┐
▼ ▼ ▼
RULES SUPERVISED ML ANOMALY ML
│ │ │
└─────────────────┼─────────────────┘
▼
RISK FUSION
│
▼
FRAUD RISK SCORE
│
┌─────────────┼─────────────┐
▼ ▼ ▼
MONITOR ALERT ACTION
This represents a transition from static detection toward adaptive behavioral intelligence.
I believe AI engineering should be supported by measurable experiments.
Typical evaluation includes:
- Accuracy
- Precision
- Recall
- F1-score
- ROC-AUC
- Confusion matrices
- Precision-Recall curves
- Feature importance
- Model training time
- False-positive analysis
For fraud detection, accuracy alone is insufficient.
A model that achieves high accuracy while missing significant fraud can be operationally ineffective.
Understand the problem before choosing the model.
Compare algorithms rather than assuming one is best.
Understand why the model produces a prediction.
Turn research into a usable system.
Treat every deployed system as an opportunity to generate new research questions.
- Supervised fraud classification
- Telecom fraud benchmarking
- XGBoost experiments
- Random Forest experiments
- Anomaly detection experiments
- AI cybersecurity analytics
- A2P/SMS fraud analytics
- Autonomous AI experimentation
- Explainable AI with SHAP
- Concept-drift detection
- Real-time fraud scoring
- Streaming fraud analytics
- Graph-based fraud detection
- Temporal behavioural modelling
- Hybrid rule + ML detection
- Human-in-the-loop fraud investigation
- LLM-assisted fraud analysis
- Multi-model risk fusion
My AI work builds on more than 20 years of telecommunications experience across revenue assurance, fraud management, billing integrity and telecom technology.
Experience includes work involving major telecom environments including:
- stc Bahrain
- Vodafone Qatar
- Etisalat / Ufone Pakistan
Areas of professional expertise include:
- Revenue Assurance
- Fraud Management
- Telecom Billing
- Roaming
- Interconnect
- Wholesale
- Rating & Charging
- Fraud Management Systems
- AI/ML Fraud Analytics
This combination of domain expertise + AI engineering is central to my current research direction.
EMBA Coursework — NUST
Bachelor of Computer Software Engineering Foundation University
AI Labs — AI Software & Research
For research, collaboration and professional opportunities:
Email: adnan_malik_83@hotmail.com
LinkedIn: linkedin.com/in/adnan-malik-30240928