Isuru Wijesinghe

Research Interests

  • Machine Learning
  • Data Mining
  • Information Retrieval
  • Medical Image Analysis with Deep Learning
  • Linear & Nonlinear optimization
  • Computer Vision

Projects

Autonomous Diabetic Retinopathy Analysis through Deep CNN Feature Extraction

Built a prediction model to classify severity level of Diabetic Retinopathy using retinal images through an ensemble of deep CNNs and then extended this classification model to a content-based image retrieval model architecture to search the collections for retinal images that have characteristics similar to the case(s) of interest.

Tools & Tech: Deep Learning, Transfer Learning, Python, Keras, Flask, Jupyter Notebook, Angular.js

Anomalies Classification in Gastrointestinal-Tract through Endoscopic Imagery with Deep Learning

Built a prediction model to classify eight-class anomalies of the digestive tract diseases using endoscopic images through an ensemble of deep CNNs.

Tools & Tech: Deep Learning, Transfer Learning, Python, Keras, TensorFlow, Jupyter Notebook

Predicting Socioeconomic Status of Sri Lanka Using Mobile Network Big Data

Built a prediction model for socioeconomic Status classification of geographical regions in Sri Lanka using anonymous caller detail records covering the voice call records, SMS records, GPRS records, and reload behaviour of millions of people in Sri Lanka and validated findings using the census data processed and published by the Census and Statistics Department of Sri Lanka.

Tools & Tech: Data Mining, Hadoop, Hive, java, R, SPSS, D3.js

Automated Computer Aided ECG System

The aim of this project is to develop a cheap and reliable ECG system to make the cost-effective. Wireless patient monitoring has become a more established technology and a natural step in this progress is to develop a reliable ECG system that contributes to the cable reduction in medical and physiotherapy environments. The ECG sensor system can connect to an ECG monitoring network using Blue-tooth technology. With the use of a microprocessor the analogue signal is digitally converted at a specific sample rate that based on the resolution of the ECG-signals.

KDD Cup 2014 - Predicting Excitement at DonorsChoose.org

project done for Data Mining module. This was held in together with DonorsChoose.org an online charity that helped poverty-stricken students, schools and teachers in America with resources from donation. We built a model to predict whether a proposed project was likely to get a donation from the donors. This was useful because rather than giving donors large number of projects to choose, gave them a set of selected projects which will save time and the likelihood of a donor donating.

Tools & Tech: R, R-studio

Sentiment Analysis on Movie Reviews

Kaggle project done for Machine Learning module. Classified the sentiment of sentences from the Rotten Tomatoes dataset which was a corpus of movie reviews used for sentiment analysis. Label phrases on a scale of five categorical values: negative, somewhat negative, neutral, somewhat positive, and positive.

Tools & Tech: R, R-studio

Transportation Management System

Developed a Mobile web-based solution for the passengers in Colombo District to reduce costs through better route planning, load optimization and mode selection.

Tools & Tech: PhoneGap, Android, PHP, JQuery mobile, HTML 5, CSS, MySQL, Google API