Predictive Algorithm Identifies Infants at Risk of Low Cognitive Ability at School-Age
A new study has developed a predictive algorithm that can identify infants at risk of low cognitive ability (LCA) by the time they reach school-age. The algorithm, created using population-based cohort data, aims to provide an early assessment of cognitive abilities in infants to help identify those who may require additional support in their educational journey.
The study, which utilized data from the Growing Up in Ireland (GUI) Infant Cohort, a survey of infants and their primary caregivers, trained various machine learning models to predict LCA at the age of 5. The models were then tested on an independent unseen test set to assess their effectiveness. Performance measures were compared, and the most significant features for prediction were identified.
To assess cognitive ability in the children, two core subtests from the British Ability Scales (BAS) Early Years Battery Second Edition were administered. The tests measured verbal ability and non-verbal ability, providing a comprehensive evaluation of the child’s cognitive skills. The scores were then adjusted for difficulty and age to ensure accurate assessments.
In determining a cut-off point for LCA, the study examined both a 1 and 1.5 standard deviation (SD) below the mean. For the main study, a 1.5 SD cut-off was used to classify children as having low cognitive ability (LCA). An additional analysis using a 1 SD cut-off was also conducted and referred to as below average cognitive ability (BACA).
The dataset used for the study contained over 600 variables across various domains, such as health, education, cognitive development, social class, and neighborhood characteristics. To select the most relevant features, Pearson correlation coefficients were calculated and plotted, and redundant features were identified and removed. Recursive feature elimination (RFE) was performed, resulting in three feature sets for the final models.
To address the imbalance in the dataset, a Synthetic Minority Oversampling Technique (SMOTE) was utilized during the training phase. The training dataset was rebalanced using SMOTE, which creates synthetic examples of the minority class to ensure better predictive performance for both classes.
Multiple machine learning algorithms, including random forest, logistic regression, and support vector machine, were trained and evaluated. After comparing the accuracy and performance across ten-fold cross-validation, the most appropriate algorithm was selected. The final models were then tested on an independent test set to assess their overall performance using metrics such as the area under the receiver operating curve (AUROC).
The study also examined the feature importance using the permutation method. Feature importance plots were created, showing the impact of each feature on the model’s performance. Partial dependence plots (PDPs) were used to visualize the relationship between the most important features and the outcome, taking into account the average effect of other features.
This research contributes to early identification and intervention for infants at risk of low cognitive ability. By utilizing a predictive algorithm based on population-based cohort data, it provides a valuable tool for educators and healthcare professionals to identify infants who may require additional support to achieve their cognitive potential. Early intervention can significantly impact a child’s future academic success, allowing for targeted interventions and personalized educational strategies.
The findings from this study highlight the importance of early cognitive assessment and the potential of machine learning algorithms in identifying infants at risk of low cognitive ability. By leveraging population-based cohort data, researchers can gain valuable insights into the factors that contribute to cognitive development and identify at-risk individuals more effectively. As technology continues to advance, predictive algorithms like these hold great promise for improving educational outcomes and supporting children’s cognitive development.
