Few-shot classification with prototypical neural network for hospital flow recognition under uncertainty

Accurately identifying and analyzing patient and personnel flow patterns within healthcare facilities is crucial for optimizing operational efficiency and delivering high-quality healthcare services. In this study, we propose a Prototypical Neural Network (PNN) tailored for few-shot learning, which effectively learns a representation space from limited labeled data. This enables efficient recognition of distinct characteristics within hospital flow footprints, ensuring examples from the same class are proximate while those from different classes are distant. Additionally, we introduce a synthetic sampling technique (SST) to address uncertainties and variations inherent in hospital personnel flow, thereby enhancing the robustness and performance of our flow recognition system. Through extensive simulation studies, we evaluate our approach and compare it against various classification methods, including support vector machine (SVM), random forest, naive Bayes classifier, residual neural network (ResNet), and fully connected neural network. The results showcase the superior performance of the proposed method, achieving an impressive accuracy of 99.17% in hospital flow footprint recognition. This outperforms classical methods, which range from 40.27% for fully connected neural networks to 80.55% for CNN. These findings underscore the efficacy of our method in recognizing hospital flow footprints, particularly in contexts characterized by uncertainty and variability.