# STUDENT PROJECTS

Following projects have been made as a part of my academic journey in groups of 2-5 members.

## MATHEMATICAL MODEL PYTHON PACKAGE

Course Work: Programming for Data Science
(M.S. in Applied Mathematics)

• Created a python module to develop and analyze mathematical models using Pandas, Matplotlib, NumPy, and Math libraries.

• Implemented the Lotka-Volterra and Chemostat model, with the scope of adding more models in the future.

## BRAIN CT HEMORRHAGE SEGMENTATION AND CLASSIFICATION USING NEURAL NETWORKS

Course Work: Machine Learning and Statistical Theory
(M.S. in Applied Mathematics)

• Trained a CNN in order to classify the location of hemorrhage in CT Scans with an accuracy of 68% on the testing set.

• Utilized U-Net, a pre-trained model to train a segmentation network with an accuracy of 92% on the testing set.

## DYNAMICAL SYSTEMS FOR GROWTH OF POPULATION OF WHALE

Course Work: Mathematical Methods and Modeling
(M.S. in Applied Mathematics)

• Developed a dynamic system to predict the population growth of whales.

• The logistic Model was also deployed on the same database to optimize the model.

• The model was studied and compared by changing parameter values.

## PULSE DIAGNOSIS MATHEMATICAL MODEL USING MACHINE LEARNING

Course Work: Final Thesis
(B.Tech Biomedical Engineer)

• Decision Tree Learning, SVM Kernel Method, and Artificial Neural Network were used to quantify the relationship between arterial pulse signal features and the presence of disease in a subject.

• Processed raw signals, developed filters to reduce noise in the collected signals.

• Due to the nature of the data, a nonlinear decision function was chosen over a linear decision function.

• On the validation set, accuracy was 73% for ANN, 55% for SVM, and 45% for decision tree.

## DEEP LEARNING FOR DIABETIC RETINOPATHY

Course Work: Enigma (Project Competition Event) in Technoczar (Technical Festival)
(B.Tech. in Biomedical Engineering)

• Deployed a CNN classifier on an image database from Kaggle to determine the presence of Diabetic Retinopathy.

• Implemented a data augmentation method and created a generator pipeline in the Keras module, in order to overcome the small magnitude of training data.

• On the validation set, accuracy was 70%​