To analyze monthly patterns and influencing factors of perinatal mortality in Ghana and compare forecasting performance of ARIMA, BPNN, DLNN, and GRNN models.
Approach:
Study Design: Retrospective, hospital-based time series analysis of 192 monthly observations from January 2010 through December 2025.
Data Collection: Data included monthly statistics on live births, stillbirths, neonatal deaths, and antenatal care indicators.
Statistical Analysis: Stationarity assessed using Augmented Dickey Fuller test; forecasting performance compared across four models.
Key Findings:
Overall perinatal mortality rate (PMR) was 24.98 per 1,000 births, based on 46,108 live births and 1,152 perinatal deaths.
ARIMA model (3, 0, 0) showed the best out-of-sample accuracy with RMSE of 11.74.
Higher antenatal care (ANC) coverage associated with lower PMR; higher hypertension burden associated with higher PMR.
Interpretation:
Perinatal mortality has declined over the long term but remains unstable; ARIMA is the most accurate forecasting model among those evaluated.
Limitations:
Neural network model design lacked hyperparameter tuning and essential architectural details.
Comparative rigor between ARIMA and neural networks is limited due to modeling shortcomings and data constraints.
Conclusion:
Monthly PMR surveillance may assist in monitoring service quality and operational forecasting.
by Agyei Helena Lartey, Denis Dekugmen Yar, Ama Asamaniwa Attua, Godfred Nyanney, Akuffo Samuel Tete Manukure, Isaac Takyi Boahen, Theophilus Oduro Kankam, Collins Mawuli Bakudie
