A Parameter Estimation Approach in Simulated Neural Data Using Metamodelling Approaches
Abstract
This study uses a metamodelling approach to compare two different sampling approaches when applying the Random Forest algorithm. In the sampling, normal sampling, and synthetic minority oversampling technique (SMOTE) with imbalanced labeled data were compared. Results show that, when using normal sampling the classification gives low prediction accuracies on the test data close to random chance. Using random forest with imbalanced labeled data and SMOTE provided an improvement on the performance of the model across classification metrics, with an average accuracy of 80%, and precision and f1 score with a similar 80% average value. A Parameter Estimation Approachin Simulated Neural Data UsingMetamodelling Approaches