Modelling Adult Obesity in China From 1975 to 2016: A Comparative Analysis of Growth Curves and Time Series Approaches

Key findings from this study

• The study found that the Logistic model provided the most accurate representation of obesity trends in China from 1975 to 2016, with near-perfect fit metrics.
• The authors report that ARIMA models proved insufficient for capturing nonlinear obesity dynamics despite adequate short-term forecasting ability.
• The researchers demonstrate that combining interpretable parametric models with adaptive deep-learning architectures creates a hybrid framework applicable to chronic disease surveillance and projection.

Overview

This research evaluated four forecasting models to project adult obesity prevalence in China across four decades (1975–2016) using World Health Organisation data. The study compared growth curve approaches (Gompertz and Logistic) with time series methods (ARIMA and LSTM) to assess their capacity to capture obesity trends amid rapid urbanization, economic expansion, and dietary transformation. Model accuracy was quantified using R², mean squared error, root mean squared error, and sum of squared errors.

Methods and approach

The researchers applied four distinct modeling frameworks to historical obesity prevalence data from China. The Logistic and Gompertz models represent parametric growth curve approaches designed to capture S-shaped and asymmetric progressions respectively. ARIMA provides a statistical time series method suited for sequential forecasting, while LSTM networks employ deep learning to capture temporal dependencies. Five-fold cross-validation validated LSTM performance. Performance metrics directly compared model fit quality across the historical period.

Results

The Logistic model achieved superior accuracy with R² of 0.9997 and root mean squared error of 0.0312, effectively capturing the sigmoidal trajectory of obesity in China. The Gompertz model attained R² of 0.9961, demonstrating strong performance in representing asymmetric long-term dynamics. ARIMA (R² = 0.8098) performed adequately for short-term predictions but failed to represent nonlinear patterns characteristic of obesity progression. LSTM achieved R² of 0.8199 with root mean squared error of 0.0582, demonstrating robust temporal learning capacity validated through cross-validation. The analysis revealed continuous, nonlinear obesity increase throughout the study period without evidence of saturation.

Implications

Obesity prevalence in China exhibits sustained growth rather than plateau, indicating ongoing susceptibility to environmental and economic drivers. The trajectory mirrors the nation's structural economic changes and lifestyle shifts, suggesting obesity will remain a significant public health burden without targeted intervention. The absence of saturation signals critical windows for prevention and control policy implementation.

Scope and limitations

This summary is based on the study abstract and available metadata. It does not include a full analysis of the complete paper, supplementary materials, or underlying datasets unless explicitly stated. Findings should be interpreted in the context of the original publication.