Performed Sythentic Minority Over-Sampling Technique to balance the dataset, performed feature engineering, developed and tuned hyperparameters for a random forest model. For the detection of Hyperthyroidism. Balancing the need for precise and accurate detection with intentional feature engineering to limit costs of unecessary data collection.
Developed a neural network model to perform multiclass sleep stage classification using seven physiological time-series signals. The project applies signal processing techniques for feature extraction and trains a deep learning model to accurately detect sleep states.
This project implements a multiclass emoji detection model that predicts the most appropriate emoji from text input. After benchmarking traditional models, we built a Bidirectional LSTM network to better capture word order and context. The final model uses text vectorization, embeddings, and a softmax output layer to classify text into six emoji categories.
Built an image classification model for breast cancer detection using transfer learning with the ConvNeXtV2 architecture. Benchmarked performance against a custom CNN and a Vision Transformer, using grid search for hyperparameter tuning. The final model leverages pretrained features to improve accuracy and generalization on medical imaging data.