Prioritizing Hospital Incident Reports Through Text Classification

منو

" Prioritizing Hospital Incident Reports Through Text Classification "
Flood, Max Madathil, Sreenath C.

Document Type	:	Latin Dissertation
Language of Document	:	English
Record Number	:	1052370
Doc. No	:	TL51487
Main Entry	:	Flood, Max
Title & Author	:	Prioritizing Hospital Incident Reports Through Text Classification\ Flood, MaxMadathil, Sreenath C.
College	:	State University of New York at Binghamton
Date	:	2019
Degree	:	M.S.
student score	:	2019
Note	:	99 p.
Abstract	:	This research explores the potential for natural language processing and binary text classification to identify and prioritize medical incident reports. The study addresses the concerns of how to efficiently process incident reports while detecting cases that yield potential for patient safety initiatives. Over 30,000 incident reports were used to build a prediction model in order to classify each incident as requiring a formal investigation or not. The prediction model implemented 10 folds cross-validation, TF-IDF bag of words, chi-squared feature selection, random under sampling, and multinomial naïve Bayes to avoid overfitting due to the data being heavily imbalanced. The model was optimized towards recall scores rather than precision or accuracy. Classification results were: accuracy = 69%, precision = 65%, recall = 84%, roc = 78%. Results showed that with basic techniques it is possible to apply text classification to medical incident reports.
Descriptor	:	Electrical engineering
	:	Health care management
	:	Information science
Added Entry	:	Madathil, Sreenath C.
Added Entry	:	State University of New York at Binghamton