Faiad, Yasmine; (2020) The Role of Early Environmental Factors in Predicting Adolescent Psychopathology: A Secondary Data Analysis of NICHD SECCYD Data using Machine Learning. Doctoral thesis (D.Clin.Psy), UCL (University College London).

Preview

Text Faiad_10116414_Thesis_volume 1_sig_removed.pdf Download (4MB) | Preview

Abstract

Background: Emotional and behavioural problems in adolescence often persist into adulthood. The early identification of those at risk of developing these problems is crucial for successful prevention efforts. A myriad of child, parent and contextual characteristics have been examined in relation to these outcomes, mostly using traditional statistical techniques. Aims: The aim of the current study is to extend previous research by using a less conventional analytic approach called Machine Learning for the prediction of two response (outcome) variables, namely externalizing and internalizing problems at 15 years of age. The examined features (predictors) encompassed variables representing characteristics of the child and his/her early infancy/childhood caregiving/interpersonal environment as well as various contextual risk factors. Methods: A secondary analysis of the data (N = 1364) from the National Institute of Child Health and Human Development Study of Early Child Care and Youth Development (NICHD SECCYD) was conducted. A set of supervised Machine Learning algorithms were used to train several predictive models and select the one with the best performance. The 5-fold cross-validation scheme was used to train and then test the model’s performance making sure that different partitions of the data are used for the training and testing phases. Embedded feature selection using Automatic Relevance Determination (ARD) was implemented to identify the features which are most relevant to the prediction of the response variables. Results: Gaussian Process Regression (exponential GPR) models had the best performance with the lowest RMSE and highest R-Squared values with respect to the prediction of both response variables. Eighteen and 11% of the variance in externalizing and internalizing problems were explained by the obtained models, respectively. The most influential features for the prediction of externalizing problems were 36-month attachment (ambivalent and insecure-controlling/insecure-other or disorganized classifications) and gender. The most influential features for the prediction of internalizing problems were 36-month attachment (ambivalent and avoidance-insecure classifications), non-family childcare hours, and ethnicity. Conclusions: The current findings are to some extent consistent with previous research. The low amount of variance explained in externalizing and internalizing problems indicate that there are other important predictors that were not included in the models. Moreover, further research is needed to test the models obtained on previously unseen data to be able to determine their clinical utility.

Download activity - last month

Export as XMLCSVJSON

Download activity - last 12 months

Export as JSONXMLCSV

Downloads by country - last 12 months

Export as CSVXMLJSON

Archive Staff Only

View Item