A Multi-Step Prediction Method Based on Small Sample Data Augmentation to Assess Wheat Flour Safety Risk.

Area of Science:

• Food Science and Technology
• Data Science and Machine Learning
• Public Health

Background:

• Food safety is critical for human health, necessitating accurate risk assessment, especially with limited detection data.
• Existing data augmentation and prediction models struggle with gradient vanishing and long-term dependency capture for food safety.
• Wheat flour safety risk assessment requires robust methods to handle data scarcity and predict future trends.

Purpose of the Study:

• To develop a Small sample Data Augmentation Multi-step Prediction Method (SDAMPM) for assessing wheat flour safety risks.
• To improve the accuracy and reliability of food safety prediction models using limited datasets.
• To provide decision-making support for reducing food safety incidents related to wheat flour.

Main Methods:

• Enhanced time-series generative adversarial networks (GANs) with temporal convolution and Wasserstein distance to augment wheat flour hazard factor data.
• Development of a dietary exposure evaluation system for wheat flour using augmented data for multi-step prediction.
• Construction of a stable Informer-based multi-step prediction model (Stainformer) with ProbSparse self-attention and dilated causal convolution.

Main Results:

• Augmented wheat flour detection data closely matched the distribution of the original limited data.
• The Stainformer model effectively predicted long-term safety risks associated with wheat flour consumption.
• The SDAMPM approach demonstrated superior performance compared to existing methods in experiments.

Conclusions:

• The proposed SDAMPM effectively addresses data scarcity in food safety risk assessment.
• The method provides accurate long-term safety risk predictions for wheat flour.
• This approach offers valuable technical support for regulatory bodies to mitigate food safety incidents.