Temporal Modeling of Cervical Cancer Risk Factors Using LSTM Networks: A Longitudinal Perspective
G. Sajiv, Natarajan Meenakshisundaram · 2025
Cervical cancer continues to pose a significant worldwide health concern, mainly avoidable yet hindered by delayed diagnosis and insufficient comprehension of changing risk factors. This research introduces an innovative time-series modeling technique utilizing Long Short-Term Memory (LSTM) networks to elucidate the temporal dynamics of cervical cancer risk variables. The model utilizes longitudinal patient data to understand how variations in behavioral, demographic, and clinical characteristics affect the results of screening tests, including Pap smears, biopsies, and Hinselmann evaluations. This study compares LSTM with Gated Recurrent Units (GRU) and attention-based architectures, illustrating that modeling temporal dependencies markedly improves early prediction of high-risk profiles. The LSTM model achieved notable classification performance, with accuracy scores of$\mathbf{9 4 \%}$on Biopsy, 95 % on Hinselmann, 97 % on the overall temporal model, and 90 % on Schiller test data. Corresponding weighted$F 1$-scores were$0.91,0.93,0.96$, and 0.85, respectively. SHAP values are utilized to enhance interpretability by visualizing the temporal significance of features. The findings endorse the creation of individualized screening protocols and adaptive risk evaluation frameworks, presenting a promising avenue for preventive cervical cancer management.