Structural Deficiency Filling Method Based on RTSDP-AE and Its Application in the Field of Agricultural Product Market Circulation

In the task of assessing the biological risk level of codling moth (Cydia pomonella) in the market circulation field using data-driven methods, the presence of multiple different dealers during the circulation process of agricultural products results in limited data collection by these dealers. This, in turn, leads to structural missing data. Unlike ordinary random missing data, structural missing data cannot be directly filled in by interpolation methods, posing a significant challenge to the risk level assessment task. This paper proposes an Autoencoder model based on regression tree similarity distribution partitioning (RTSDP-AE). This model can identify data in the complete dataset that has a similar distribution to the missing variables of the incomplete data, and use this data to reconstruct the missing variables of the incomplete data to fill in the missing parts. The study is currently conducting a risk assessment on 208 batches of apples being transported from Gansu to Beijing. It has been found that the assessment performance of the RTSDP-AE model proposed in this paper is generally superior to that of other risk assessment models.