数据挖掘的特征介绍和处理方法介绍

青旅半醒 2021-11-01 11:03 950阅读 0赞

** 一、特征工程**

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70][]

二、数据清洗：

1、数据样本抽样

2、异常值（空值）处理

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 1][]

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 2][]

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 3][]

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 4][]

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 5][]

在特征预处理的过程中，进行数据标注是非常重要的，比如我们想要预测明天下雨不下雨，那么这个时候我们得到的数据中，需要对关于我们预测的结果的数据进行标注，也就是下雨不下雨进行标注。

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 6][]

上图是一个hr的人力资源的分析图，我们用这个数据会去预测员工是否离职。 并且建立模型，得出什么样的人会离职。

而我们的labe标注在left上面。属性为1为离职，属性为0代表不离职。

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 7][]

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 8][]

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 9][]![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 10][]

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 11][]

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 12][]

**特征变换：对指化，就是对特征进行对数化和指数化。**

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 13][]

先对特征进行指数化，然后再进行归一化（softmax）。

** 对特征进行对数化：**

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 14][]

数据离散化：就是将连续数据变成离散数据的过程

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 15][]

1.  方法就是分箱技术：等频分箱，等距分箱
2.  朴素贝叶斯算法：需要的数据就是离散的

### **                                                                       分箱技术：** ###

**根据深度和宽度对数据进行分箱**

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 16][]

第一个是等深分箱，和等距分箱的实践![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 17][]

归一化是数据在\[0,1\]之间：![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 18][]

** 正规化：用在每个对象的各个特征的表示最多**

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 19][]

** 用python操作正规化：**

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 20][]

根据上面的公式得出的正规化的数据

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 21][]

*** 注意下面的是 L1,L2的正规化。***

from sklearn.preprocessing import Normalizer
    import numpy  as np
    #正规化实际上是对行进行正规化的
    
    x = Normalizer(norm="l1").fit_transform(np.array([1,1,3,-1,2]).reshape(-1,1))
    print(x)
    b = Normalizer(norm="l2").fit_transform(np.array([[1,1,3,-1,2]]))
    print(b)

正规化的结果如下：

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 22][]

### *** 特征降维：*** ###

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 23][]

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 24][]

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 25][]

** 用python操作lda的实现过程**

import  numpy as np
    from sklearn.discriminant_analysis import LinearDiscriminantAnalysis
    X = np.array([[-1,-1],[-2,-1],[-3,-2],[1,1],[2,1],[3,2]])
    Y = np.array([1,1,1,2,2,2])
    m_lda = LinearDiscriminantAnalysis(n_components=1).fit_transform(X,Y)
    print(m_lda)

lda实现结果：

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 26][]

[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70]: /images/20211101/5ecb261881874f8cb5ae012da25b693d.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 1]: /images/20211101/269d00db49784f2890e6094128566340.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 2]: /images/20211101/407fe396cc85426f814c8bc4bb67f663.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 3]: /images/20211101/ca6ca7e6223a43a2ab942be943e0de14.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 4]: /images/20211101/66876f4321224bd99d96ebad10dcc583.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 5]: /images/20211101/e7e283a74c574127840e236193156470.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 6]: /images/20211101/1295da0cc2744ff183bf64e67a217e83.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 7]: /images/20211101/815d6281ac434b1699fc4018d0e4177a.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 8]: /images/20211101/d739824a91c0445d91d34e670cad5372.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 9]: https://img-blog.csdnimg.cn/20190811110435405.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 10]: /images/20211101/d0c7ad2a44cd44568f6e7cb7c3ddac17.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 11]: /images/20211101/a78659f9e4ed4d4d8b7e132ad12a02f1.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 12]: /images/20211101/9439eef8659e4bbe9f6322ee904d696b.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 13]: /images/20211101/f9e201fddb9b444b84eb0757b4e60c2c.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 14]: /images/20211101/aad1ca538f2a483484ba538520080175.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 15]: /images/20211101/275d01a16828438c8ec9310510903679.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 16]: /images/20211101/cb4760c08ae3458591e929af44e93981.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 17]: /images/20211101/100470f8a0ff4007803a5d8797533f3f.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 18]: /images/20211101/96dd68f8669143059bf57046a1b1e340.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 19]: /images/20211101/3122ccdc23dd4e279fbb417d2f080c70.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 20]: /images/20211101/12efc9ddecb8489883c83b4c70c19c70.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 21]: /images/20211101/2cd08449d1a3418f90ba68484486900c.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 22]: /images/20211101/4d08b3bfb38f4d69b6ec5923ad71053e.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 23]: /images/20211101/389c5b8a5fd440d59eb6d3f876b7ea1c.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 24]: /images/20211101/4d6dbd1496ae4317aaed9b14577c5b34.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 25]: /images/20211101/2ab167ecd2914737b0d713859f263b00.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxNDc5NDY0_size_16_color_FFFFFF_t_70 26]: /images/20211101/209aabf1fd9a4b95964a5de766e9a515.png