简体繁体 English

如何提高svm

[英]How to improve svm

原文 2019-05-29 10:39:21 3 1 python/ svm/ data-science

I apologize in advance for my bad English. 对于我的英语不好，我事先表示歉意。

So, I'm working on my final year study project and the description says to make a state of the art on the methods often used to predict churn in telecommunications and then choose two methods that will be applied to the data. 因此，我正在研究我的最后一年的研究项目，并且描述说要对经常用于预测电信用户流失的方法进行最先进的研究，然后选择将应用于数据的两种方法。

It also says to try to add my contribution to one of the methods. 它也表示尝试将我的贡献添加到其中一种方法中。

I chose the decision tree and SVM methods. 我选择了决策树和SVM方法。 I would like to add my contribution to the SVM method but I do not know how. 我想对SVM方法做出贡献，但是我不知道如何做。 I did some research and the most common thing is the "Cross-Validation" method but since it is used by everyone, is it considered a contribution? 我进行了一些研究，最常见的方法是“交叉验证”方法，但是既然每个人都在使用它，那么它是否被认为是一种贡献？

I also thought about a hybridization but I'm not sure which algorithm would be best for that. 我也考虑过杂交，但是我不确定哪种算法最适合。

So I wanted to know if you could give me some ideas to explore in order to try to improve this algorithm, whether in precision, speed or otherwise. 因此，我想知道是否可以给我一些思路，以尝试改进该算法，无论是精度，速度还是其他方面。

If I sound like a beginner, that's because I am XD. 如果我听起来像个初学者，那是因为我是XD。

1 个解决方案

I am also a beginner in this field , but i can give you some pointers i've come across, 我也是该领域的初学者，但是我可以给我一些提示，

You can look at newer feature generation(Try to do research on that fields specific to telecommunications) 您可以查看更新的功能（尝试在特定于电信的领域进行研究）
Use different algorithm for imputation(KNN, central imputation). 对插补使用不同的算法（KNN，中央插补）。
If you want high accuracy go for XGBOOST . 如果要高精度，请使用XGBOOST。
As this is a churn problem i would concentrate on the recall. 由于这是一个客户流失的问题，我将集中讨论召回问题。