Data Science Portfolio by Sumiko

Data Science Projects

Multi Modal Skin Disease Diagnosis Model

• Research project on multi-modal computer vision (VGG-16) and natural language (BERT) model to diagnose skin diseases of 4 classes from images and textual symptoms achieving 88% accuracy on validation set exceeding single-modal models performance
• Dataset is from Dermnet (images) and text symptoms dataset is web-scraped using Beautiful Soup

NUSecure Security Monitoring Proof of Concept

• An integrated proof of concept using AWS EC2 featuring a Flask dashboard app for incident monitoring, a mySQL database and machine learning pipeline for time series forecasting future incident rates and case prioritisation

American Sign Language Computer Vision Application

• NUS Orbital Project involving computer vision, machine learning and visual design to translate and teach American Sign Language to a greater audience.

Predicting concrete strength using regression

• Data Science project from research, data cleaning, exploratory data analysis for prediction of concrete strength using regression methods

Credit card fraud detection using classification

• Data Science project from research, data cleaning, exploratory data analysis for prediction of credit card fraud using classification methods

Classification Problems

Credit card default classification using neural network and PCA

• Practice classification problem on credit card default using neural network classification and exploring different model specification and PCA

Convolutional neural network for object identification

• Using convolutional neural network to do image classification on fashion MNIST

Regression Problems

LASSO Regression using auto-MPG dataset with hyperparameter tuning

• Practice regression problem on classic auto-MPG dataset with exploration of hyperparameter selection using cross-validation

Simple regression on diabetes dataset

• Predicting glucose levels using simple linear regression with simple residual analysis