IKE, CHIEMERIE GIBSON (2026) ANOMALY PREDICTION MODEL FOR CHILD DEVELOPMENT (DELAYED SPEECH) USING MACHINE LEARNING. Other thesis, Godfrey Okoye University, Enugu.
|
Text
IKE_CHIEMERIE_GIBSON (1)_Phoenix.pdf Download (8MB) | Preview |
Abstract
Childhood developmental anomalies such as speech delay, ASD, and ADHD account for an estimated 10–15% of children worldwide, but in Nigeria, most cases go unnoticed until the affected child reaches school age, owing to the critical lack of skilled developmental experts and excessive dependence on clinical expertise rather than diagnostic tools. The current study sought to design a machine learning-based intelligent model called the Anomaly Prediction Model for Child Development that is able to predict childhood developmental anomalies within the age group of zero to five years based on simple clinical and demographic features. After an exhaustive analysis of seven peer-reviewed journals within the years 2016 to 2025, it was determined that Random Forests, Support Vector Machines, and Logistic Regression have proven their effectiveness in structured development prediction problems with questionnaires providing accuracy rates of 70%-90%. A two-step design approach of Object-Oriented Analysis and Design (OOAD) in conjunction with CRISP-DM (Cross Industry Standard Process for Data Mining) was utilized where OOAD was responsible for application layer design and UML modelling, and CRISP-DM provided structure for the 6 step ML pipeline.Publicly available data sets from the developmental milestone database from the CDC, standards for child development from the World Health Organization, and data sets found in published research studies have been obtained and pre-processed using the programming language Python (using Pandas) to form a representative training data set. Data processing involved feature engineering, handling missing values, normalization, and an 80/20 split between training and test sets. Several supervised machine learning algorithms were then trained using Scikit-learn including Random Forest, SVM, and Logistic Regression. Random Forest method generated improved performance results on classification accuracy measures which outperform all benchmarks created by other researchers. Feature selection identified developmental milestone deviation, medical history of patients, socio-demographic factors, and behavioral factors as critical factors for diagnosis. The system architecture design involved the creation of a framework which included a presentation layer, application layer, and database layer based on SQL based relational database management systems. This study shows that the construction of a reliable, affordable, and scalable prediction system for developmental disorders is possible through use of freely available information and open source software applications.
| Item Type: | Thesis (Other) |
|---|---|
| Subjects: | Q Science > Q Science (General) |
| Divisions: | Faculty of Engineering, Science and Mathematics > School of Electronics and Computer Science |
| Depositing User: | MICHAEL MADUBUKO |
| Date Deposited: | 22 Jul 2026 11:37 |
| Last Modified: | 22 Jul 2026 11:37 |
| URI: | http://eprints.gouni.edu.ng/id/eprint/5998 |
Actions (login required)
 |
View Item |
