Abstract Misinformation about child development can misguide parents and educators, influencing early learning practices and beliefs. This study applies data science and probabilistic modelling techniques to distinguish myths from facts within child-development discourse. A curated dataset was compiled from verified educational and health websites and validated by domain experts. Several probabilistic and ensemble models—GaussianNB, MultinomialNB, BaggingNB, XGBoost, and the proposed Multinomial Distribution-based Bayesian Model–Extended (MDBM-X) —were evaluated across multiple feature representations, including Bag-of-Words (BoW), TF-IDF, and word embeddings. MDBM-X refines the traditional Multinomial Naïve Bayes (MNB) framework through optimised smoothing (= 0. 1), bigram BoW representation, and chi-square-based feature selection (6500 terms). Using stratified ten-fold cross-validation and 1000 bootstrapped resamples, MDBM-X achieved a mean accuracy of 95. 7% (95% CI 94. 5–96. 9%) and a single-split peak of 98. 6%. Statistical testing confirmed that its improvement over baseline models was non-random and moderately strong (t (9) = 2. 25, p 0. 05, Cohen’s d = 0. 71). Compared to transformer-based baselines (BERT-base), MDBM-X delivered comparable accuracy with substantially lower computational cost and higher interpretability. The findings demonstrate that carefully optimised Bayesian models can serve as lightweight, transparent, and statistically reliable frameworks for misinformation detection in sensitive domains such as child development, supporting evidence-based decision-making for parents, educators, and policymakers.
Tajrian et al. (Mon,) studied this question.