Built a university Pattern Recognition project that compares classical and machine-learning classifiers through feature selection, PCA/LDA transformations, hyperparameter tuning, and cross-validated evaluation on a real binary classification dataset.

Independent CPU reproduction of ICML 2026 #31878 'Instance-Level Costs for Nuanced Classifier Evaluation' (NEC metric) on real Jigsaw/NHANES/Turkey data. Challenge judge: quality high.

BigQuery SQL workflow for estimating classifier recall using stratified sampling and weighted evaluation.

The project focuses on analyzing neural activity data to classify neuron types (spiny and aspiny). It integrates unsupervised learning methods (PCA, Autoencoders) and supervised learning models (Logistic Regression, MLP) to build accurate classifiers that effectively analyze neurons' electrical responses.