Moving towards accurate and early prediction of language delay with network science and machine learning approaches