Network Perception Of Campus Spatio-Temporal Data Analysis

With the widespread use of mobile smart devices,Wireless Local Area Networks(WLANs)are facing thousands of connection requests every day,and these data imply macro and micro perspectives on individual behavior patterns to conduct student behavior.This article uses the self-developed platform WiCloud to obtain the spatiotemporal data of WiFi connections of students on the campus of Beijing University of Posts and Telecommunications,and conducts research from the following three aspects:network-aware data mining paradigm,correlation analysis of campus student behavior and academic performance,and campus student movement Prediction of behavioral intentions.1.Based on the uniqueness of WiFi connection data,summarize the mining paradigm of such data.Information mining paradigms such as terminal Mac address,RSSI signal strength,and connection timestamp contained in such data belong to the process of progressively describing individual activity patterns based on multi-granularity.There are four main steps.The first step is to formally describe in the dimensions of time and space,and to clarify the conversion relationship between time granularity and different granularity.The second step is to mine the areas where individuals reside frequently,marking the individual's access to the area.,Analyze its temporal characteristics,the fourth step is to form a sequence of individual activities,and then find observations and verify them.2.Based on the WiFi connection,a network awareness system(WiCloud system)collects student terminal access records,serializes the connection behavior of a single individual in a time series according to a certain size window,and obtains a behavior vector that marks a single individual within a certain time,and then uses Mathematical function that maps a vector to a scalar that can be quantified and compared.At the same time,students' actual academic performance maps are used as labels,and classic machine learning classification algorithms are used to train and predict individual data and features.In horizontal comparison,linear classifiers such as logistic regression and support vector machines perform better than non-linear classification.Devices such as GBDT and NB,in which the LR accuracy rate is as high as 0.98,which verifies the validity of the method and the calculated features have certain predictive power for the prediction result,that is,the academic performance,and provides a lower cost for student behavior prediction and analysis on campus More explanatory methods.3.Model the individual mobility process in the campus scene as a Markov Decision Process(MDP)process.Based on the reinforcement learning framework,model the campus physical environment and individual students as an environment in reinforcement learning.When the agent is triggered,the agent predicts the next movement of the student,thereby forming a decision sequence.The reinforcement learning agent interacts with the environment through trial and error,and learns an optimal strategy over time to understand deeply.The intention of the individual's movement behavior is to optimize the long-term benefits and the accuracy of the prediction during the multiple rounds of interaction between the agent and the experiment.The experiments show that the method proposed in this paper is effective for predicting the intention of students moving on campus.A stable positive return was obtained during the experiment.The normalized return after the experimental results converged remained constant above 0.6 and the prediction accuracy was average Greater than 0.8.