Using the TSA-LSTM two-stage model to predict cancer incidence and mortality

Area of Science:

• Oncology and Public Health
• Data Science and Machine Learning

Background:

• Cancer is a leading global cause of death, with lifestyle factors like smoking, obesity, and lack of exercise significantly contributing to incidence and mortality.
• Accurate correlation analysis and prediction of cancer trends are crucial for public health guidance and resource allocation.

Purpose of the Study:

• To enhance the accuracy of cancer incidence and mortality prediction by analyzing lifestyle correlations.
• To develop and validate an optimized prediction model for future cancer trends.

Main Methods:

• Data preprocessing involved sample expansion using cubic spline interpolation, converting annual data to monthly data for improved analysis.
• A Two-Stage Attention Long Short-Term Memory (TSA-LSTM) model was developed, incorporating input feature attention and time performance attention for enhanced prediction.
• Factor analysis was employed to identify key lifestyle factors influencing cancer development.

Main Results:

• The TSA-LSTM model demonstrated high accuracy (98.96%) in long-term cancer trend projection.
• Analysis confirmed strong correlations between lifestyle factors (e.g., smoking, obesity, alcohol abuse, lack of exercise) and various cancer types.
• Visual tests indicated a direct link between poor lifestyle and increased risk for lung, laryngeal, oral, breast, colorectal, and colon cancers.

Conclusions:

• Lifestyle modifications are critical for cancer prevention.
• The TSA-LSTM model offers a robust tool for accurate cancer incidence and mortality prediction.
• Cancer incidence is projected to increase significantly in the coming years, underscoring the need for proactive public health interventions.