Article Figures & Data
Figures
- Figure 1.
Data preparation flowchart through to model selection (full image is available at: https://github.com/kapoor1992/spinal_cord_injury_recovery/blob/release/submission/src/ml/modelling/plots/flowchart.png).
- Figure 2.
Top 20 most important features (full image is available at: https://github.com/kapoor1992/spinal_cord_injury_recovery/blob/release/submission/src/ml/modelling/plots/importance.png).
Tables
- Table 1
Feature mappings from those in the original dataset to transformed versions used at model training time, along with the imputation techniques used
Original feature Machine learning features Missing value imputation technique Occupation status—injury Homemaker, in training, in workshop, other, retired, student or infant,
unemployed, unknown, workingMapped to unknown Diabetes—history Same as original feature Mode (no history) Veteran Same as original feature Mode (not a veteran) Race Asian, black, multiracial, Native American, unknown, white Mapped to unknown AIS—admission A, B, C, D None; rows dropped Sex Same as original feature None; all values were populated Education—injury Same as original feature Mode (high school) Depression—history Same as original feature Mode (no history) TBI likelihood—injury Same as original feature Mode (improbable) Level of injury—admission Same as original feature None; rows dropped Daily alcohol—history Same as original feature Mode (zero) Anxiety—history False, general anxiety, multiple, PTSD, panic disorder, unknown Mapped to unknown Primary insurance True, false, unknown Mapped to unknown Age—injury Same as original feature None; all values were populated Loss of memory—injury Same as original feature Mode (no loss of memory) Marital status—injury Divorced, living unmarried with partner, married, never married,
other, separated, unknown, widowedMapped to unknown Neurologic category—
admissionComplete paraplegic, complete tetraplegic, incomplete paraplegic,
incomplete tetraplegic, minimal deficit paraplegic, minimal deficit
tetraplegic, unknownMapped to unknown Loss of consciousness—
injurySame as original feature Mode (no loss of consciousness) TBI, Traumatic brain injury.
Description Value Total samples 20,790 Training samples 18,737 Training injury dates 1972–2005 Testing samples 2053 Testing injury dates 2006–2016 Original features 18 Machine learning features 53 - Table 3
Accuracy results from machine learning model runs, ordered in descending order by test accuracy
Model Train accuracy Test accuracy Ridge Classifier 0.824 0.736 SVM 0.825 0.735 Elastic Net 0.823 0.732 Logistic regression 0.824 0.732 Ensemble (Elastic Net, KNN,
Random Forest)0.860 0.717 CNN 0.840 0.711 Random Forest 0.927 0.693 Naive Bayes 0.591 0.429 Rank Machine learning feature 1 AIS—admission_A 2 AIS—admission_D 3 AIS—admission_B 4 Neurologic category—admission_complete paraplegic 5 Neurologic category—admission_incomplete tetraplegic 6 AIS—admission_C 7 Neurologic category—admission_complete tetraplegic 8 Neurologic category—admission_incomplete paraplegic 9 Level of Injury—admission 10 Marital status—injury_never married 11 Marital status—injury_married 12 Occupation status—injury_working 13 Primary insurance_unknown 14 Occupation status—injury_student or infant 15 Age—injury 16 Marital status—injury_divorced 17 Race_white 18 Sex 19 Occupation status—injury_retired 20 Education—injury
Extended Data 1
Spinal cord injury recovery-release-submission. Download Extended Data 1, ZIP file
In this issue
Spinal Cord Injury AIS Predictions Using Machine Learning
