PRANAV B R

Data Analyst with proven expertise in developing fraud detection pipelines utilizing machine learning algorithms at Codec Technologies. Skilled in cloud computing and data visualization, delivering actionable insights through data analysis and statistical modeling. Strong problem-solving abilities and collaborative approach drive impactful business decisions.

1. Anomaly-Driven Financial Fraud Detection System:
Tools & Tech: Python, Scikit-learn, Pan

das, SMOTE, AutoEncoders, Isolation Forest, Power BI, Tableau  
I designed and implemented a system to detect anomalies using Isolation Forest and AutoEncoders. This system identifies fraudulent transactions. I addressed class imbalance with SMOTE and undersampling. I also developed a real-time dashboard to show high-risk transactions and send fraud alerts.

2. HR Analytics Dashboard for Employee Attrition Prediction:
Tools & Tech: Python, Scikit-learn, Seaborn, Plotly, Logistic Regression, Random Forest, Tableau, Power BI  
I built a model to analyze and predict employee attrition using HR datasets. I identified key factors such as age, salary, and tenure. I trained classification models and created an interactive dashboard to monitor attrition KPIs and assist HR in decision-making.

3. AI-Driven Supply Chain Risk Intelligence Platform with Predictive Disruption Alerts  
Tools & Tech: Python, Scikit-learn, NLP, XGBoost, Time Series Analysis, Power BI/Tableau  
Developed an AI platform to monitor and predict supply chain disruptions using historical, real-time, and external data like weather and news. Used NLP and time series models to create predictive risk alerts. Built a visual dashboard to track supplier performance and notify stakeholders.

4. Real-Time Hyperlocal Demand Forecasting and Dynamic Resource Allocation System  
Tools & Tech: Python, ARIMA, LSTM, Geospatial Analysis, Pandas, Plotly, Power BI  
Created a real-time system to forecast hyperlocal demand using ARIMA and LSTM models, integrating geospatial and temporal data. Enabled dynamic resource allocation across locations based on predicted demand spikes. Visualize results through an interactive dashboard for operations teams

5. 3D partially occluded face recognition using hybrid techniques 

Tools & tech: Python, OpenCV, MATLAB, 3D morphable models, PCA, CNNs 

I designed and built a 3D face recognition system that can accurately identify faces even when they are partially covered by masks or sunglasses. This system uses a mix of traditional methods and deep learning techniques. I used benchmark datasets like FRGC v2.0, BU-3DFE, FaceWarehouse, and Bosphorus for training and testing the model. I applied principal component analysis (PCA) to reduce dimensions, and performed 3D face alignment for normalization. CNNs helped extract texture-based features The system showed strong accuracy in recognizing faces and worked well in real-world situations with occlusions It has potential uses in biometric authentication, security surveillance, and smart access control systems

Research Publication:
“A Survey on Multimodal Approaches for 3D Face Recognition in Occluded Environments”
International Journal of Advanced Research in Computer and Communication Engineering (IJARCCE) – Vol. 14, Issue 6, June 2025
DOI: 10.17148/IJARCCE.2025.14655 |Impact factor: 8.471 published as part of B.E. Final year project Proposed a hybrid deep learning framework integrating ResNet50, PointNet++, and transformer-based attention mechanisms for robust 3D face recognition under occlusions, such as masks and sunglasses Demonstrated enhanced performance using multimodal feature fusion and real-world benchmark datasets (BU-3DFE, FRGC v2.0, Bosphorus).