College Sports Decision-Making Algorithm Based on Machine Few-Shot Learning and Health Information Mining Technology.

Few-Shot Learning has had a significant influence on how people live, work, and learn. Physical education is a requirement for a college diploma. Sports management systems, which focus on data collection, organization, and analysis, as well as timeliness and guidance, are one of the current challenges in the field of physical education at the country's top colleges and universities. The amount of sex in the room is minimal. Time is money when it comes to making college sports decisions, and this paper uses data from physical fitness tests to illustrate this point. Use Few-Shot Learning technology to extract relevant data from the data, allowing teachers to provide more scientific and effective guidance and suggestions to students. The design and implementation of this paper collect data from physical fitness tests in real-time using mobile edge computing, analyze the data, and display the results using machine learning technology, which mines deep features and displays analysis results, can be used to evaluate students' physical fitness. The data and information in the physical fitness analysis system are more readable and time-saving, allowing students to better understand their true level of physical fitness. Because of the results of data mining, teachers can provide more specific guidance and recommendations for each student's physical characteristics.

1. Introduction

With the deepening of the practical application and theoretical research of big data, big data has become a new growth point of the economy and society, as well as an emerging analytical tool for humanities and social science research, which has put forward innovative requirements for decision support of college sports in the era of big data [1]. The rapid development of big data and the Internet, cloud computing, and other technologies have prompted the growth of data in the information age, and the accumulation of data has become more and more massive, with the consequent generation of industry data in massive areas. Sports can improve people's physical and mental health, as well as guide them toward developing scientific and healthy lifestyles and habits and promoting their overall development. It is conducive to deepening sports reform, enabling the sports industry to be developed more systematically and comprehensively, and enhancing the vitality and momentum of the sports industry's development; it is conducive to increasing employment opportunities and raising the employed population; it is conducive to deepening sports reform, enabling the sports industry to be developed more systematically and comprehensively, and enhancing the vitality and momentum of the sports industry's development. The main shortcomings of such tools are manifested in the manual way of input work intensity; form stacking complex, inefficient, weak analysis, data readability is poor. Traditional physical education adopted statistical tools, often taking Excel office software and comprehensive class education office system. It not only adds to administrators' and teachers' daily workloads but it also makes data processing and analysis more difficult. Daily teaching activities and physical fitness tests cannot provide students with real-time and effective feedback [2]. In terms of analysis methods, the analysis methods used in the above tools are still stuck on simple variance, mean, and reliability calculations, resulting in the conclusions obtained from the analysis staying on the surface and failing to bring the full value of a large amount of data into play. However, the data and conclusions that administrators, teachers, and students care about are precisely these less easily discovered and valuable hidden information. To meet the development needs of big data analytics and to facilitate the deep integration of big data and its applications in various fields, research on big data analytics process modeling techniques is conducted to fully consider the ease of data analysis, domain complexity, and efficiency of execution of big data analytics. To improve the efficiency of big data analysis, to allow users to focus on domain business analysis logic rather than tool usage in big data analysis, to establish a domain-oriented reusable, well-structured processing framework for big data analysis processes, and to rely on the Hadoop platform with distributed storage scale and parallel computing capabilities. This aids big data analysis and value discovery in a variety of fields, and it is critical for developing scalable and user-friendly big data intelligent analysis software systems. The integration of existing resources and the development of a professional university sports decision support system can solve problems and issues encountered on the road to sports development, provide advice and strategic support for the development of sports, and accelerate the construction process of physical education in the context of globalisation and informatization [3].

In this paper, I attempt to use data mining techniques to study and analyze data on college students' physical fitness, develop a new college sports decision support system, and mine and analyze data on college physical fitness in order to uncover more valuable hidden information. This will assist students in better understanding their physical fitness status and teachers in initiating appropriate teaching activities on time. It improves the quality of university physical education by making physical education more timely, effective, and relevant; it also assists students in improving their physical fitness and developing good exercise habits.

2. Related Work

With the in-depth research of data mining technology, the application of data mining technology has been gradually extended to different fields, and some of the scholars have applied data mining technology to the education industry.

The literature [4] systematically discusses the importance of data mining techniques applied to education and teaching. The literature [5] employs crude sugar set theory and mining analysis to determine the relationship between improved student performance and active learning IM's dominant factors. There is still a lot of data in the university education system that can be mined, such as teaching evaluations, student performance, student information, and so on. However, data mining research in college education is currently primarily theoretical, and there are few shaped products that apply to data mining technology. The number of students and data managed by colleges and universities is growing year by year, and the manual processing mode in dealing with student information and student achievement can no longer meet the current needs. More scholars and researchers are applying data mining techniques to college education and physical ability analysis in this environment. Literature [6] uses the classification method of a decision tree, applies data mining technology to the student performance information, and constructs a professional ability decision tree model to help teachers gain more accurate and efficient insight into the problems that exist in the teaching process, and achieve the effect of optimizing teaching quality by using performance information. The literature [7] adopted the decision tree ID3 algorithm, and the association rules Apriori algorithm for data mining analysis based on student performance data. The ID3 algorithm was analyzed to get which factors are related to students' good grades; the association rules Apriori algorithm was analyzed to dig out the degree of influence of course excellence on other courses. The literature [8] used the FP-Growth algorithm to study the student physical fitness test data from a deeper level based on the physical fitness test data from six colleges and universities. The results showed that nearly half of the students in the six colleges were not at the standard weight and the results of the algorithm run observed that the students lacked training for lower body strength in their physical training and had significantly weaker lung capacity rating and endurance rating, suggesting that the students should strengthen their aerobic training. The literature [9] used the association rule Apriori algorithm to filter out five strong association rules about male and female students, respectively, based on the physical test data of college students in a university. The results showed that under the condition of “total score = pass,” more female students failed in the test item standing long jump, and more male students failed in the test item pull-up. This identifies sports that need further attention in the future at the university to strengthen the overall physical fitness of students. The literature [10] used student data from a US university to develop an early warning system for students. The literature [11] proposes a method for predicting students' knowledge of the necessary skills for their majors by mining information from their academic performance once the students' bias in the learning process has been compensated for. Literature [12] uses the Hadoop big data platform to mine the data of informative campus applications in order to recommend campus information based on learning characteristics. The literature [13] examines the innovation of graduate physical education's deep evaluation mechanism in the era of big data from three perspectives: student-oriented value logic, practical logic with the goal of theory implementation, and problem-oriented reality logic. The literature [14] examines the general idea of reforming the evaluation method of college physical education based on big data from a theoretical perspective before delving into the specific application of big data evaluation on a practical level. The literature [15] begins with the current state of operation and growth trend of the big data platform for physical education in colleges and universities, analyses its goals and values, and investigates its architecture and characteristics in order to develop an excellent scheme to promote the construction, operation, and management of the big data platform for physical education.

It can be seen that a large number of research findings on big data have aided in the development of quantitative analysis methods in physical education. These findings seize big data's innovation opportunity and examine the innovation of quantitative analysis methods of physical education promoted by big data from multiple perspectives, based on a comprehensive analysis of big data's characteristics and mining its value; However, the majority of big data and physical education research is based on a general perspective of physical education or a more general study on the overall grasp of physical education, and the relevance of theoretical guidance, as well as the operability of the application of theoretical results, must be improved. Simultaneously, research into the integration and innovation of big data and quantitative analysis methods in physical education is primarily based on theoretical or empirical feasibility analyses, with the realistic path of how to apply it not being sufficiently explored.

3. Research and Construction of a Decision Support System for University Sports Based on Big Data Analysis Technology

3.1. Big Data Analytics Technology

In the current network era, the value of data as the core of big data does not arise out of thin air and requires the help of certain mining techniques.

Data mining usually has the following eight steps: (1) information collection: abstract the analyzed object, get the characteristic information of the analyzed object, and use reasonable information collection methods to load the characteristic information of the analyzed object into the database. (2) Data integration: the information collected from various objects is centralized to facilitate the subsequent correlation analysis work. (3) Data statute: The data volume after data integration is generally large and difficult to handle. The data set obtained can be statutorily represented, and the data volume after the statute is much smaller than the original data, but still maintains the integrity of the original data, and the data mining result after the statute is the same as a result before the statute. (4) Data cleaning: Some of the data collected in the database may be incomplete, noisy, or have conflicts, etc., so data cleansing work is needed to improve the data in the database and eliminate all kinds of illegal data. (5) Data transformation: the data in the database will be transformed into a form that is convenient for data mining, and the common ways and means are smooth aggregation, data generalization, etc. (6) Data mining: the data information stored in the data warehouse, the use of appropriate analytical tools and statistical methods, processing data, and finally getting the analysis results. (7) Pattern evaluation: Validate the data mining results from the business perspective and analyze whether the data mining results are correct. (8) Knowledge representation: presenting data mining results to users, which can use various visualization tools, reporting tools, etc.

Deep learning [16-18] is one of the key technologies for tapping the value of data at the level of big data; at the level of deep learning, the collection and use of a large amount of data has the potential to improve the accuracy of machine models. The meanings of TP and TN when classifying a class in a machine model both represent the case where the classification result is correct: TP is a positive class, and TN is a negative class. FP denotes that the incorrect category is divided into the correct, while FN denotes that the correct category is divided into the incorrect. As a result, the support of big data is required to develop machine learning, and the assistance of machine learning is required to mine the value of big data, and the two are mutually reinforcing and interdependent. The study's main algorithm is machine learning, which is a broad term for a specific type of algorithm. Machine learning algorithms [19, 20] attempt to intervene or classify a large amount of raw data in order to uncover hidden laws in the data and discover the data's value, allowing data models to be built. As shown in Figure 1, it mainly consists of three types, and this study is chosen to implement model building with supervised learning.

Figure 1

Machine learning classification graph.

Concerning the machine models in this study, they are all practically similar to decision tree classification models, and all amount to a combination of multiple dichotomous classification problem models. Therefore, before proceeding to model generation, the first task is to gain a detailed understanding of the binary classification problem. The next example will be picture content discrimination: there is a picture stored in a certain computer gallery. The task of the computer at this point is to determine the specific content in the current picture. After making the relevant determination, the computer has to check whether the judgment output meets the expected effect. Therefore, a feature vector X can be set in the computer for representing the result. Then, the current computer can use the computer language Y = 0 or Y = 1 to indicate the right and wrong results of the judgment, the main formula used as shown below:

CRM, as a cascade regression model, usually requires first integrating the predictions of each decision tree and calculating the average as the final prediction; however, the classification problem employs a different strategy: the plural voting method, which requires counting the number of votes received for each type of label and selecting the one with the most votes. As the final prediction result, the classification labels are output. According to the previous analysis, some factors tend to influence the combined classifier's generalization ability, which is mainly related to the classification performance of individual metaclassifiers from the perspective of individual metaclassifiers; from the perspective of the set of metaclassifiers, the size of the correlation between metaclassifiers is also an important influencing factor.

The boundary function of a decision forest is described as follows:

The categorical efficacy of decision forests is defined as follows:

Define the original prime function of the decision forest as rm(θ, x, y), and the boundary function as the mathematical expectation of the original prime function on m(x, y), whose expression is as follows:

Since the mathematical expectation of the variance satisfies the following:

Therefore, it can be obtained that the generalization error of the decision forest PE satisfies the inequality

The decision tree's definition and principle have been explained. The decision tree is essentially a process of growing data from the root node to the leaf nodes in a continuous split. The branching direction of nodes is currently determined in the construction of decision trees primarily based on the judgment criteria of the classified nodes, which cannot be changed once the direction has been established. Following VGGNet, 3 ∗ 3 small convolution kernels became the standard for network design. Most general neural networks start with a 7 ∗ 7 large convolution and then use 3 ∗ 3 convolution stacking (ResNet, DenseNet), whereas lightweight neural networks start with a 3 ∗ 3 convolution and then use 3 ∗ 3 convolution stacking. This method can make the network classify the best at the current node, but it cannot guarantee the best final classification result, according to the Info-Gain principle. A “probabilistic” decision tree is designed to balance the current optimum and the final classification result. This tree does not determine the split of each node but rather describes it in a probabilistic manner. Because the number of labeled samples in practical recognition tasks is usually limited, using ISM for target detection tasks has some drawbacks.

Figure 2 explains the structural framework of the decision tree algorithm, where the voting codebook is obtained by sliding windows and extracting feature channels (feature descriptions), and in the training process, a series of decision trees are formed by supervised target training and heuristic algorithms, and further formed into a decision forest; in the target detection process, the voting codebook obtains numerous target locations by weighted voting, and the greedy algorithm is used to solve for the maximum target probability to obtain the final target location.

Figure 2

The structural framework of the decision tree algorithm.

3.2. Research and Construction of a Decision Support System for University Sports Based on Big Data Analysis Technology

Colleges and universities provide physical education in the form of college physical education courses for physical exercise, education, and guidance. The issue is primarily focused on the fact that physical education in China is still a teaching system centered on physical education teachers and that the quality of physical education is influenced to some extent by the disparity in physical education teacher quality. Simultaneously, the physical education system's and statistical tools' backwardness lowers the quality of teaching and learning in three ways: first, the traditional method of performance evaluation and statistics makes teachers' work tasks cumbersome and inefficient; second, the traditional method of performance evaluation and statistics makes teachers' work tasks cumbersome and inefficient. Second, the single system for evaluating physical education results and physical tests makes it difficult to apply the results to each student's problems of physical fitness and health. Third, based on the above two points, the heavy workload of teachers makes it difficult to give students effective guidance advice immediately; the single sports assessment makes it difficult to come up with professional guidance and feedback. For the above problems, data mining methods can be adapted to analyze the sports test data of college students and the whole process. After determining the content and purpose of the project, then the relevant data are collected and preprocessed, where the data preprocessing includes data selection, data cleaning, data integration, and data specification, in four steps. Finally, data mining processing is then performed on the data set using relevant data mining models to obtain data mining results. The actual content of the project is linked to getting the corresponding value knowledge. The process is shown in Figure 3.

Figure 3

Data mining process for student physical tests in universities.

The personal data analysis module can realize the detection and analysis of personal physical fitness test data and judge the comprehensive evaluation of individual users and fitness methods. Test indicators include three major indicators of body shape, body function, and physical quality. Morphological indicators mainly include waist circumference, scapular skinfold thickness, height, hip circumference, weight, chest circumference, abdominal skinfold thickness, and upper arm skinfold thickness; functional indicators mainly include step index, veins, systolic blood pressure, vital capacity, and diastolic blood pressure. Qualities mainly include choosing reaction time, push-ups, back strength, grip strength, sitting forward bends, sit-ups, vertical jumps, standing on one foot with eyes closed. Secondly, adults are grouped by age and gender, each group is 5 years old, and there are 16 groups of men and women in total. 10 test indicators are used to analyze individual physical fitness test results, which is convenient for the rapid processing of physical fitness test data. Potential value analyze the needs of individual users and managers, provide specific implementation methods for each module, and complete the overall design of the service application platform.

After performing machine learning, the result obtained from the learning is often an optimal decision tree, which leads to the establishment of a comprehensive evaluation model of college students' physical fitness. Next, the performance of the model needs to be evaluated using the sample data of the physical fitness test. Only when the evaluation result is good using the sample data of the physical fitness test, the comprehensive evaluation model of university students' physical fitness has the most “optimal” set of functions to solve the problem.

The personalized fitness mode recommendation service solution based on machine learning mainly includes three parts: the collection of basic user data, the discovery of fitness modes that match personal fitness test data and interest needs, and the recommendation of personalized fitness modes that match personal fitness test information and so on. For the generation of the fitness mode recommendation model based on machine learning, machine learning is then mainly applied to tap the value of personal fitness test information and carry out fitness mode recommendations suitable for individual users' fitness test information. Using the fitness test data uploaded by the administrator or the personal fitness test data stored in the database by the service application platform as input, the fitness test data is classified by the machine learning algorithm after entering the model, and the corresponding fitness mode categories are extracted from the fitness resource library, and the fitness modes of the corresponding categories are called through the Web UI module in the recommendation list and presented to the individual user through the service application platform presented to individual users.

The scale transformation mechanism makes the variable scale cluster analysis method capable of automatic execution, ensuring that the decision-maker is almost not required to participate in the method execution process, i.e., the decision-maker only needs to subjectively select the type of scale transformation strategy and deterrence. This method can directly obtain all college students' physical test results that satisfy decision preferences and their performance on the appropriate analysis level, assisting the university sports director in making the best sports training decision for the students. Because the traditional cluster analysis method can only perform cluster analysis at a single scale level, and the management business itself has multilevel characteristics, i.e., competition support groups made up of various participating students each have their own appropriate levels of decision analysis, the traditional cluster analysis method can only perform cluster analysis at a single scale level. All three scale levels have unsatisfactory classes in the single-scale clustering results. Although unsatisfactory classes are always present in the traditional cluster analysis method's clustering results, as the scale hierarchy grows, the number of unsatisfactory classes decreases significantly. This is strong evidence that scales can link decision-making activities to subjective and objective data and that scale transformation can add more valid information and knowledge to operational data, reducing decision complexity.

In addition to the basic scale hierarchy, the variable scale cluster analysis method can obtain more accurate scale characteristics of satisfaction classes than the traditional cluster analysis methods at the same level. Since the scale transformation process of variable scale cluster analysis starts from the basic scale level and takes the lowest scale level of each satisfaction class as its appropriate decision analysis level, the satisfaction class consistency theorem ensures that the clustering results are consistent among different data analysis levels so that each satisfaction class retains its most detailed scale characteristics, allowing analysts to have more accurate information and thus improve the quality of decision results [21]. For the traditional clustering analysis method, even if it can find the partial satisfaction class in the process of single-scale data analysis, it does not guarantee that the scale characteristics of the satisfaction class reach a better analysis level at this time, which leads to the problem that the result class characteristics are often not significant in solving practical problems in management.

4. Experimental Verification and Conclusions

Clustering validity can be evaluated in terms of internal validity evaluation and external validity. Since there is a scale transformation iterative transformation process in the execution of the variable scale clustering analysis method, and the set of objects (thesis domain) decreases with the increase of the number of iterations, there is a situation that the results of variable scale clustering analysis are distributed in different analysis levels. The internal validity index can only evaluate the quality of clustering results on the same analysis level, so to ensure the objectivity and fairness of the evaluation results; this paper adopts the external validity evaluation method to verify the variable scale clustering analysis method. The experiments are conducted on the premise of satisfying the methods and principles of multiscale data model construction and randomly generating multiscale data models, and the main purpose is to conduct relevant experimental analysis on the validity and parameter sensitivity of the variable-scale clustering analysis method. Since the experiments take an external validity evaluation index to test the clustering effect, data labels are added to this data model. The basic task of the experiments on the validity of the variable-scale cluster analysis method is to test the clustering effect of the variable-scale cluster analysis method compared with the traditional single-scale cluster analysis method. Because the meta-partitioning cluster analysis algorithm's initial class center selection is random, this paper first repeats the class analysis work 50 times on each of the 54 single-scale data models obtained from the multiscale data model and then takes the mean, maximum, minimum, and standard deviation of the results of these 50 experiments, as well as the maximum value, minimum value, and standard deviation to complete the full-scale spatial clustering analysis. The variable scale clustering analysis method's validity is tested by comparing its results to those of the traditional single scale clustering analysis method at the basic scale level to see if the variable scale clustering analysis method can meet the clustering results' quality requirements. The results of the clustering validity index evaluation between the variable-scale cluster analysis method and the traditional single-scale cluster analysis method at the optimal scale level are compared to see if the variable-scale cluster analysis method can improve clustering analysis efficiency. The clustering validity index evaluation results are shown in Figure 4.

Figure 4

Results of the evaluation of clustering validity indicators.

Sensitivity analysis of satisfaction determination thresholds for variable scale cluster analysis methods. The basic task of the experimental analysis of satisfaction determination thresholds & sensitivity for variable scale cluster analysis methods is to test the clustering effect of variable scale cluster analysis methods under the conditions of different satisfaction determination thresholds. It is the maximum granularity deviation taken from all the basic scale clustering results that satisfy the business requirements. In this experiment, the clustering effect of the variable scale clustering analysis method under all possible values is examined by setting the experimental analysis range of the satisfaction determination threshold parameter. As can be seen from Figure 4, among the 50 clustering experiments of the variable scale cluster analysis method, the average clustering validity results of the variable scale cluster analysis method for all the above evaluation indexes are better than the average validity results achieved by the traditional single scale cluster analysis method at the basic scale level, and the quality improvement rate of the clustering results is more than 10%, among which the quality improvement rate of the results under the NMI evaluation index reaches 1. 5. 63%. 5.63%, proving that the variable scale clustering analysis method can better meet the validity requirements of the clustering results of the traditional single scale clustering algorithm.

According to the basic task of the sensitivity experiment of the variable scale cluster analysis method's satisfaction threshold R, the experiment compares the results of the external validity evaluation indicators of the variable scale cluster analysis method under different satisfaction threshold R0 and investigates the degree of influence of the satisfaction threshold R 0on the validity of the variable scale cluster analysis method. To investigate the relationship between satisfaction determination thresholds and the effectiveness of the variable scale clustering analysis method and to compare the trends of the clustering effect of the variable scale clustering analysis method under various satisfaction determination thresholds, the satisfaction threshold R0, which was determined subjectively by the analyst, was 3.8 in the above-mentioned experiments on the effectiveness of the variable-scale cluster analysis method. The satisfaction threshold's experimental analysis range was set to a region around 3.8, which was specifically set to [3.0, 6.0] in this experiment, and the step of change was 0.1. Each satisfaction threshold was subjected to 50 runs of the variable-scale cluster analysis method. The cluster result validity evaluation indicators were averaged to perform a satisfaction determination threshold sensitivity analysis.

Figure 5 shows the results of the sensitivity analysis of the parameters of the scale cluster analysis method. By comparing the experimental results, it is found that the variable scale clustering analysis method is less affected by the satisfaction determination threshold R0 and the results have stability, as discussed below: although the results of all evaluation indicators of the variable scale analysis method fluctuate in the range of the satisfaction determination threshold parameter, the maximum fluctuation does not exceed 1%, indicating that the validity of the clustering results of the variable scale clustering analysis method is not sensitive to the satisfaction determination threshold. The overall trend of the evaluation index results of the variable scale cluster analysis method increases slightly with the increase in the value of the satisfaction determination value parameter, indicating that the overly strict initial satisfaction constraint is not conducive to the optimal solution of the variable scale cluster analysis method.

Figure 5

Results of sensitivity analysis of parameters of the scale cluster analysis method.

After obtaining valuable data based on machine learning decision trees and multiscale cluster analysis methods, the data can be applied to the actual curriculum for decision support. Classroom monitoring is mainly achieved through wearable sports bracelets, which can monitor various indicators such as student exercise intensity, exercise density, heart rate curve, and heart rate warning. The sports bracelet has four colors: green, blue, orange, and red, and it can be used to guide teaching by monitoring students' heart rates. When the red logo appears, it means the student's heart rate is higher than normal, and the sports watch will sound an alarm, alerting the teacher to reduce the intensity and volume of exercise. The teacher can then use big data to analyze the movement of students through the class's background and determine whether the class has met the teaching objectives. The use of the decision support system by teachers is depicted in Figure 6.

Figure 6

Actual use of the decision support system by teachers.

According to the research, 94.4 percent of students believed the decision support system could achieve targeted teaching, while 5.6 percent believed it could not. This means that the majority of physical education teachers still believe that decision support systems can help them deliver more targeted instruction and learning. At the same time, the decision support system can make intelligent statistical analyses of students' learning performance, as shown in Figure 7. Based on the conversion scores of students' indicators, and statistical comparison and analysis based on items, gender, grade level, and other degrees, the system can continuously track students' physical development trends, quickly grasp students' physical and athletic qualities, and conduct targeted teaching according to students' differences, to solve the problems of students' “not being able to eat” and the problem of “not being able to eliminate.” And through intelligent statistics out of the student results can not only be viewed by individual students, but also for classmates, teachers, school leaders, and other views. The percentage of teachers who often use the decision support system to push PE homework to students is 35.2%, the percentage of PE teachers who use it occasionally is 53.7%, and 11.1% of PE teachers will not use the platform to push homework to students, with teachers occasionally assigning homework accounting for the largest percentage. This indicates that physical education teachers do not frequently use the decision support system to assign physical education homework to students. Physical education is now gradually being paid attention to in junior high school teaching, recognizing that cultivating students' physical quality cannot be accomplished overnight, that a few minutes in the classroom is insufficient, and that teachers must reasonably arrange students' physical education homework to ensure that students can develop good physical exercise habits whether in classroom teaching or at home. This will help junior high school students grow in a healthy way. It is a new trend in physical education to assign “physical education homework,” which will serve as a strong motivator for students to engage in physical activity. PE homework will not only encourage students to engage in physical activity after school, but it will also draw parents' attention to their children's physical development and improve their physical performance.

Figure 7

Results of intelligent analysis of learning achievement.

As can be seen in Figure 8, the use of decision support systems for learning by students grows steadily with grade level. The lowest usage rate is among the first-year students, with 24.1%. The next highest usage rate is among sophomores, with 33.2%. The highest rate of use was among third-year students, with 42.7%. The move of junior high school students to use the decision support system for learning makes it possible to deeply integrate “Internet+” with college sports. The decision support system is a manifestation of the integration of Internet+ with sports, which breaks the traditional way of learning sports. For example, the learning space has changed a lot. Physical education is traditionally taught by teachers in physical education classes, and students learn the content from the teachers. The learning space must be set up in a specific location on campus. Schools with better hardware facilities have more indoor learning space and are less affected by weather, but schools with poor hardware implementation may not be so fortunate and will be impacted by weather. When it rains, for example, physical education classes are canceled, and the physical education program in junior high schools has only one class per week, so there is not much physical education left in a semester after holidays and rainy days. Physical education teaching tasks in high schools cannot be completed on time, in a high-quality and quantity manner, and students' physical health will not improve but will deteriorate as they grow older. The advent of the decision support system allows schools to overcome learning space constraints; the transformation of students' learning spaces adds color to the traditional way of physical education learning, while the decision support system also contributes to college physical education innovation.

Figure 8

Usage statistics of the decision support system.

5. Conclusions

While data mining techniques have made significant progress in a variety of fields, the use of fitness analysis in conjunction with data mining techniques is less common. The main reason for this is the limitations of physical education in colleges and universities, where major institutions have vastly different curricula and evaluation criteria. This has resulted in a physical fitness and health management system based on data collection and statistics, making guidance and educational significance difficult to achieve. With the advent of the big data era, the problem of data support for college physical education decision-making guidance has been solved, and the “online + offline” hybrid teaching model has emerged and permeated the college physical education teaching model, which is not only innovation of the college physical education teaching model, but also an important reflection of the new curriculum concept. Furthermore, the rapid development of big data has aided the development of multifunctional school teaching instruments. Teaching can benefit from a web page with a larger capacity and faster page updates, which provides richer and more specialized information resources. The research objects in this paper are physical test data and physical health self-assessment data, and the physical test data have uniform standardization and guidance but are not strong for readability and guidance. In order to achieve better physical fitness health education goals, this paper employs data mining technology and design to implement a physical fitness analysis system. The theory of data mining is discussed, as well as data mining algorithms such as the decision tree and association rule algorithms and the decision tree C4.5 algorithm and association rule algorithm. To mine the physical fitness data information, the apriori algorithm and multiscale clustering algorithm are chosen, and the results of big data analysis are also used to design, implement, and verify the physical fitness analysis system for college students.