| 摘要: | 隨著科技的發達和數據分析的蓬勃發展，各種產業都有越來越多的大型企業嘗試運用資料探勘的方式來將銷售中所取得的大量資料轉換成有用的資訊，以求節省公司的成本或是增加利潤，而在這樣的背景之下，有許多資料探勘的工具和相關語言隨之而生，本研究主要是從眾多的資料分析工具中，選取LibSVM和LS-SVM兩種分類工具，針對兩種特徵屬性多且資料量大的資料集HIGGS和covertype做探勘後的數據比較，過程中會針對隨機抽樣不同的training data和testing data，搭配不同的kernel function和不同的SVM工具來做交叉分析，透過得出的實驗數據，評估在何種搭配下，分析的時間/正確率比例較小，可以取得較高的效益。並且使後續想運用SVM工具來做資料分析的研究者，以此為依據而針對欲分析的資料集類型取得較好的搭配效果。
本實驗結果主要列出資料分析的時間和正確率，以及時間增加/正確率增加的比例，找出時間和正確率較高，且時間增加/正確率增加的比例較低的組合，研究後發現，在LibSVM使用Linear kernel時，分析HIGGS和covertype資料集可以取得較少的時間和較高正確率，但同時在training data數提高時的效率會較低，而LS-SVM在分析兩種資料集時正確率較高但分析時間較長，且training data數提高的時候效益較LibSVM來的低。
;With the development of science and technology and the vigorous development of data analysis, there are more and more large-scale enterprises in various industries trying to use data mining methods to convert the large amounts of data obtained in sales into useful information in order to save the company’s costs. or increase profits, and in this context, there are many tools for data exploration and associated languages was invented.
This research selects LibSVM and LS-SVM classification tools from numerous data analysis tools, and compares the data from two kinds of data sets, HIGGS and covertype, compare Random sampling of different training data and testing data, with different kernel functions and using different SVM tools for cross-analysis, then through the experimental data obtained, to assess under what collocation, the analyst can get higher benefits. and for subsequent researchers who want to use SVM tools for data analysis, they can obtain better collocation effects for the type of data set to be analyzed based on this research.
The results of this experiment mainly list the time and accuracy of data analysis, as well as the ratio of the increase in time/accuracy rate. wish to find the combination which time and accuracy had better performance, and the ratio of the increase in time/accuracy rate is low. the result can be found that when using the Linear kernel for LibSVM, analyzing the HIGGS and covertype data sets can achieve less time and higher accuracy, but at the same time the efficiency will be lower when the number of training data increases.furthmore, LS-SVM gets better correct rate but the analysis time is longer, and when the training data increases, it’s efficiency is lower than the same condition of LibSVM |
