Document Type
Project
Advisors
Dr. D. Robert Adams; adamsr@gvsu.edu
Embargo Period
12-19-2018
Abstract
Systemic Lupus Erythematosus (SLE) is a chronic autoimmune disease that causes the immune system to attack the body’s own connective tissues and organs. Humans have difficulty predicting SLE symptom severity levels because of the complex interactions of disease trigger exposure levels over time. To address this issue, we constructed a novel machine learning solution that generates a model capable of predicting SLE symptom severity levels with 8.3-19.9% average error. It does so by inputting trigger exposure levels into a recursive neural network and training them with a unique method that continually turns training on and off based on the maximum error each day. This allows the RNN to learn SLE flare activity without overtraining remission activity, thus maintaining a greater degree of plasticity. Models trained in this fashion performed 3.5-5% better on average than those trained via the standard method. Future areas of work include replicating these results with a large patient training data set and evolving the model to predict disease trajectory.
ScholarWorks Citation
Skocelas, Katherine G., "Systemic Lupus Erythematosus Symptom Severity Prediction Using a Recursive Neural Network" (2018). Technical Library. 324.
https://scholarworks.gvsu.edu/cistechlib/324
