| Summary: | 碩士 === 國立台北師範學院 === 數理教育研究所 === 90 === This research mainly uses data mining to discover knowledge in the database of elementary students’ physical fitness tests. Data of this research were gathered from the elementary students aged at 11-12 years old in Taoyuan County, Taiwan in Year 2000. Their results of physical fitness tests were used as the database. The total valid sample is 41,795 students. Research tools include three data mining softwares: STATISTICA v6.0, CART® v4.0, and XpertRule® Miner v1.47. Results of this research are as the followings:
1. By using the data visualization technique, results show that student body figures were linearly correlated with their weights, waist lines, and hip sizes, except their heights. Most relevant scatter plots of waist lines and other variables illustrate a big cluster and a small cluster. Results of two items in physical fitness tests reveal significant difference among students: the biggest difference was in 800-meter run and the second was in standing long jump. It was found that 30-second sit-ups and 60-second sit-ups were positively correlated.
2. Researcher divided the database into 9 clusters for doing cluster analysis. Among these clusters, the biggest difference was found in 800-meter run and standing long jump was ranked as the second. The big cluster consisted by Cluster 1, Cluster 2, and Cluster 8 had the most members, and their physical fitness conditions were medium range. Students in Cluster 7 and Cluster 9 comparatively performed well in each test. Students in Cluster 3, Cluster 4, Cluster 5, and Cluster 6 were comparatively worse than others.
3. By adopting Twoing algorithm, researcher found a decision tree of 18 leaf nodes. The error rate of the testing set was 0.056. The most important variable is 800-meter run, and the second is standing long jump.
4. From the database, researcher found two association rules under the minimum support of 0.1 and the minimum confidence of 0.75. If replaced by BMI (Body Mass Index) and the ratio of waist line to hip size, there were four association rules founded. The association rule of ''''(136 < standing long jump < 158),( 268 < 800-meter run < 317) (physical fitness = middle)'''' has the higher confidence, approximate 80%. Although the support of this association rule is not the highest, still is more than 10%. It should be the best association rule of this research.
|
|---|
