| |
|
| |
|
|
|
| |
|
A multi-contextual examination of predictors of expressive vocabulary among late and typical talking toddlers using a machine learning approach |
|
|
|
|
|
|
|
|
|
|
|
|
|
Julia Nikolaeva - Northwestern University
|
Elaine Kwok - Northwestern University
|
Soujin Choi - Northwestern University
|
Brittany Manning - Northwestern University, Weill Cornell Medicine
|
Lauren Wakschlag - Northwestern University
|
Elizabeth Norton - Northwestern University
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| SRCLD Year: |
2022 |
| Presentation Type: |
Poster Presentation |
| Poster Number: |
PS2F20 |
| Presentation Time: |
(na) |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
There is considerable unexplained variation in expressive vocabulary in toddlerhood, as few studies consider the dynamic interplay among various domains that influence variability in developing language. The goal of this study is to identify the strongest predictors of 30-month-old parent-reported expressive vocabulary based on (1) traditional child language measures, as well as other domains hypothesized to improve prediction (2) child brain measures (electroencephalography/EEG), (3) parent-child transactional language environment, (4) family risk/protective factors, and (5) child mental health risk. Random forest machine learning methods are employed to explore linear and nonlinear patterns of data and test several models simultaneously for better prediction accuracy. In proposed analysis with existing data from 178 toddlers (50% late talkers), we will use the random forest algorithm to run a series of random decision trees to determine both the combination of and individual ranking of predictors that contribute most to overall accuracy of predicting expressive vocabulary size at 30 months of age. Predictors identified as most important will suggest areas of future research to improve early identification of language delays and disorders. Funding Source: NIDCD R01DC016273. |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|