Projects

Evaluation of different models for price prediction

Comparing different methods to predict AirBnB prices in New York City — regression, random forest, XGBoost, neural networks

Using a dataset from AirBnB Open Data, I conducted performance comparisons to predict the prices of stays. The best performance was achieved with a five Dense-layer network with regularisation and dropout with an accuracy of 70%, while XGBoost achieved 68%.

Feature engineering and machine learning clustering

Clustering customer data to identify and profile a certain type of customer that will conduct a precise banking operation

Using Santander customer data, the goal of this project was to explore different classification methods to predict a certain type of banking transaction. The best model was a combination of XGBoost and StratifiedKFold cross-validation.

Bird species classifier with transfer learning

Transfer learning applied to bird recognition

The idea of this project was to explore the difference between a model built from scratch and a pretrained model. In the field of computer vision, pretrained models require significantly less data, and without surprise the pretrained model (ImageNet18) surpassed the locally built model.

Performance assessment of option pricing methodologies

Computational analysis of different option pricing methodologies based on Monte Carlo simulation

This project explored different variations of Monte Carlo simulations. The analysis revealed that Quasi Random Monte Carlo Simulation performed the best among the variants (antithetic variables, Quasi MC, Random MC). The project used Black-Scholes European Call as a benchmark and tracked two metrics: speed of convergence and variance of results.

Facial recognition, deep learning model and statistical analysis

Computer vision and facial recognition applied to TV series Suits

I designed a face recognition pipeline using the YOLO algorithm and OpenCV to identify characters in the TV series Suits (treating the series as a sequence of images). It achieved a 97% accuracy in face recognition.