Study on Model Based on Blstm to Multi-Classify and Recognize Poverty Stricken Students in College

Abstract

The BLSTM neural networks is utilized to construct a six-layer bidirection recurrent neural network by taking advantage of its long-term memory and powerful ability of features learning. By using the data set which consist of lunch consumption of college students and actual economic situation survey results which would be used as label data. After data augmentation, The training task was carried out on it, at last, the model was extracted by parameters tuning including selecting of dropout rate, optimization algorithm, loss function, activation function etc. The model is capable of muti-classifing and recognization of poverty stricken students in college. When using this model to recognize tricken students of other classes, A more satisfied and accurate result was got comparation to HMM algorithm, the average accuracy is 90%.