In this study, we identified the cognitive parameters of rats with traumatic brain injury (TBI) in an eight-arm radial maze and used them for TBI classification through machine learning models. A total of 16 cognitive parameters were derived using a sensing trajectory bitmap in the eight-arm maze. Of these 16 parameters, five (i.e., short-term memory error, latency, total distance, frequency of movement from an arm without food to an arm with food, and frequency of entry into the arm on the right after exiting an arm) were selected as representative parameters and were input into four machine learning models, namely, support vector machine (SVM), decision tree, random forest, and k-nearest neighbor (KNN) models, to classify and compare sham rats and rats with TBI. The performance evaluation results for the machine learning models revealed that the SVM model had the best performance among the models. Its overall accuracy, sensitivity, and area under the receiver operating characteristic curve (AUC) were >85, 98, and >94%, respectively. At some postsurgical time points, the sensitivity and AUC of the SVM model even approached 100%. The random forest and KNN models had satisfactory performance on Day 28 postsurgery. Overall, the SVM model had satisfactory performance in classifying both mild and severe TBI. Our findings can serve as a reference for future research on TBI feature classification.