Predicting the cognitive ability of young women using a new feature selection algorithm

Background and aims: Cognitive abilities are the mental capabilities to perform several processesthat include: executive function, comprehension, decision-making, work performance, andeducational attainment. This study aimed to investigate the relationship between several biomarkersand individuals’ cognitive ability using various machine learning methods and a new featureselection algorithm.Method: A total of ۱۴۴ young women aged between ۱۸ to ۲۴ years old were recruited into thestudy. Cognitive performance was assessed using a standard questionnaire. A panel of biochemical,hematological, inflammatory, and oxidative stress biomarkers in serum and urine was measuredfor all participants. A novel combination of feature selection and feature scoring methodsin a hierarchical ensemble structure has been proposed to find the most effective features to recognizethe importance of various biomarker signatures in predicting cognitive abilities in youngwomen. Feature selection with different classifiers was used to construct models. In this manner,using three filter methods, the scores of each feature were considered. The union of high-scoringfeatures for each filter method was stored as the primary feature subset. Moreover, the high-accuracyfeature subset was selected by a wrapper method. ۱۰-fold cross-validation was applied andrepeated ten times to avoid random events in searching the feature subset. Afterward, the mostrepetitive features were chosen using the relative frequency of each feature in ten folds that werehigher than a defined threshold. Ultimately, if a high-scoring feature provided a high relativefrequency, it could be nominated as a significant feature. This step gives the best feature subset toreduce feature dimensions while preserving accuracy.Results: Among the ۴۷ extracted factors, the serum level of nitric oxide (NO), alkaline phosphatase(ALP), phosphate as well as blood platelet count (PLT), were entered into the model ofcognitive abilities with the highest accuracy using a decision tree classifier.Conclusion: The serum levels of NO, ALP, phosphate, and blood PLT count, may be importantmarkers of cognitive abilities in apparently healthy young women. These factors may provide asimple procedure to identify mental abilities and earlier cognitive decline in healthy adults.

کلیدواژه ها:

Algorithms ، Biomarkers ، Cognitive Performance ، Feature Selection Methods ، Machin Learning