Title :
Biological data classifications with LDA and SPRT
Author :
Nkounkou, Brittany ; Lee, Chih ; Huang, Chun-Hsi ; Brown, Colin
Author_Institution :
Comput. Sci. & Eng. Dept., Univ. of Connecticut, Storrs, CT, USA
Abstract :
We investigate classification algorithms LDA, SPRT and a modified SPRT on clinical datasets for Parkinson´s disease, colon cancer, and breast cancer. The SPRT algorithms were run with components in decreasing variance order and random order. Results for those in random order were calculated as the majority predictions over 100 runs. Truncation was always set to the total number of components of the dataset. Accuracies for each algorithm were determined using the method of leave-one-out cross-validation. The highest accuracy for the Parkinson´s disease dataset was 0.7128, generated by the MSPRT-random function at α = β = 0.175 and by the SPRT-random function at α = β = 0.32. The highest accuracy for the colon cancer dataset was 0.8871, generated by the MSPRT-ordered function at α = β = 0.22. The highest accuracy for the breast cancer dataset was 0.9648, generated by the MSPRT-ordered function at α = β = 0.07 and by the SPRT-ordered function at α = β = 0.09.
Keywords :
biological organs; cancer; cellular biophysics; data analysis; gynaecology; medical diagnostic computing; patient diagnosis; pattern classification; probability; random functions; Parkinson disease; biological data classification; breast cancer; clinical datasets; colon cancer; data analysis; leave-one-out cross-validation; linear discriminant analysis; modified sequential probability ratio test algorithms; random function; biological data classification; discriminant analysis; sequential probability ratio test; supervised learning;
Conference_Titel :
Bioinformatics and Biomedicine Workshops (BIBMW), 2010 IEEE International Conference on
Conference_Location :
Hong, Kong
Print_ISBN :
978-1-4244-8303-7
Electronic_ISBN :
978-1-4244-8304-4
DOI :
10.1109/BIBMW.2010.5703792
