Cognitive and Task Predictors of Naval Submarine School Academic Performance: A Pilot Study

Abstract

Retaining highly qualified and trained service members (SMs) is critical for maintaining the readiness of the U.S. military to execute its mission. Unplanned losses, related to SM termination before completing their first contract, harm readiness and incur unanticipated expenses. Improved prediction of a SM’s academic performance during initial skills training could improve operational outcomes by reducing SM separations related to poor grades. Cognitive assessments that evaluate skills specific to military occupational specialties may help predict training performance, yield opportunities for customized intervention, or guide the selection of SMs to jobs that match their cognitive skills and abilities. We compared three machine learning algorithms (linear discriminate analysis [LDA], K-Nearest Neighbors [KNN], and Support Vector Machine [SVM]), which classified the initial skills training scores of 22 SMs as low (score cut-off < 75%) or high (score cuff-off >85%) on five separate exams administered during military ascension training, using performance on a ten-task cognitive assessment battery. The battery measured neurocognitive domains of attention, visual learning, working memory, abstraction, and vigilance. The cut-off scores characterized the lower and upper performance range. The resulting models exhibited modest predictive capabilities in classifying academic exam performance, with recall and precision performance in the 50th and 60th percentile. Only the KNN and SVM models exhibited better-than-chance classification performance (p < .001). Separately, correlational analyses found that performance on a simulated sonar task accounted for 31% of the variance in academic performance. The findings of this study imply that future research should add these promising cognitive measures to aid in screening and help more students achieve academic success.

Keywords: sonar, academic success, cognitive battery, predictive modeling

DOI: 10.54941/ahfe1006281