Prediction Of Student Behavior Based On Student-card

The continuous growth of campus data has made it possible for us to use data to better develop students.How to integrate the information about teachers and students and help students' life on campus is a complicated issue.Improving student services and optimizing management methods is a tough job for universities.For example,School canteens need to estimate how many people eats at meals.School transportation needs to plan student activity routes.Mental health departments need to pay attention to unsociable students.Accurate prediction of student behavior is an effective way to solve these problems.We can more convenient to locate the student's position and get the status information about the student's history by the campus card.In the student card data,we collect the access controls data and consumption data of the students.These two parts of the data are related to the students' school activities.We construct the time series characteristics of student behavior based on the card data.We propose the model to predict students' future positions and consumption.And then,it gives the main study tasks and contributions of this thesis as follows.(1)Time-series based student position prediction method.Using the student's card data,we build the RNN model with the times series features in the student history card.The loss function of the model is the variant of Focal-loss.Through this model,we can predict where students will appear at some point in the future,both short-term and longterm.At the same time,we use the neural network embedding method to generate a set of vectors for each user,which can reflect the students' behavior patterns.(2)A method of predicting student amounts based on time series.Like location prediction,we add the student's consumption information to the RNN network.The structure and loss functions as the network are modified,so that we can use the RNN model to predict the amount of consumption of the future.(3)Time-series based student behavior fusion model.Due to the correlation between the position of students in the future and the amount of their consumption,they cannot be perceived in a single model.We have merged the above two models,so that the model can better forecast the position and the consumption amount of the students.In the experimental results,compared with the traditional machine learning model,our model is more effective.In the prediction of college canteen traffic,our model also provides more accurate prediction values.For campus service optimization,student traffics monitoring,this model has practical meaning.