数据挖掘

逃离我推掉我的手 2024-02-19 20:53 76阅读 0赞

分类与预测

1常见的分类与预测算法

![SouthEast][]

（1）回归分析

![SouthEast 1][]

![SouthEast 2][]

逻辑回归模型：

![SouthEast 3][]

逻辑回归模型建模步骤：

![SouthEast 4][]

逻辑回归代码：

1.  `#如下可运用于根据特征来判断违约情况等`
2.  `import pandas as pd`
3.  `#提取数据`
4.  `filename= './data/bankloan.xls'`
5.  `data=pd.read_excel(filename)`
6.  `x=data.iloc[:,:8].as_matrix()`
7.  `y=data.iloc[:,8].as_matrix()`
8.  
9.  `#建模`
10. `from sklearn.linear_model import LogisticRegression as LR`
11. `from sklearn.linear_model import RandomizedLogisticRegression as RLR`
12. 
13. `rlr=RLR()#建立随机逻辑回归模型，筛选变量`
14. `rlr.fit(x,y)`
15. `rlr.get_support()#获取特征筛选结果,为索引值`
16. `print(u'通过随机逻辑回归模型，筛选结束')`
17. `print(u'有效特征为：%s'%','.join(data.columns[rlr.get_support()]))`
18. `x=data[data.columns[rlr.get_support()]].as_matrix()`
19. 
20. `lr=LR()#建立逻辑回归模型`
21. `lr.fit(x,y)#训练模型`
22. `print(u'逻辑回归模型训练结束。')`
23. `print(u'逻辑回归模型平均正确率：%s' % lr.score(x,y))`

(2)决策树

![SouthEast 5][]

这里主要介绍用于离散分类的ID3算法：

![SouthEast 6][]

信息增益值越大，不确定性越小，相应特征属于最佳特征值

决策树算法预测销量高低：

1.  `import pandas as pd`
2.  
3.  `inputfile = './data/sales_data.xls'`
4.  `data=pd.read_excel(inputfile)`
5.  `#数据是类别签，要转化为离散数据`
6.  `#用1表示‘好’，‘是’，‘高’；，用-1来表示相反的属性值`
7.  `data[data.values==u'好']=1`
8.  `data[data.values==u'是']=1`
9.  `data[data.values==u'高']=1`
10. `data[data.values != 1] = -1`
11. `x=data.iloc[:,1:4].as_matrix().astype(int)`
12. `y=data.iloc[:,4].as_matrix().astype(int)`
13. 
14. `from sklearn.tree import DecisionTreeClassifier as DTC`
15. `dtc=DTC(criterion='entropy')#建立决策树`
16. `dtc.fit(x,y)#训练模型`
17. `#可视化数目`
18. `from sklearn.tree import export_graphviz`
19. `x = pd.DataFrame(x)`
20. `import pydotplus`
21. `from IPython.display import Image`
22. `#利用Graphviz进行可视化`
23. `x = pd.DataFrame(x,columns=data.columns[1:4])`
24. `dot_tree = export_graphviz(dtc, feature_names = x.columns, out_file = None,filled=True, rounded=True,special_characters=True)`
25. `graph=pydotplus.graph_from_dot_data(dot_tree)`
26. `img=Image(graph.create_png())`
27. `graph.write_png('out.png')`
28. `#建立好tree.dot后，需要在Graphviz中将其转化为img`

（3）神经网络

决策树算法预测销量高低：

1.  `import pandas as pd`
2.  `#参数初始化`
3.  `inputfile = './data/sales_data.xls'`
4.  `data = pd.read_excel(inputfile,index_col=u'序号')#导入数据`
5.  `#数据是类别标签，要将其转换为数据`
6.  `#用1来表示‘好’‘是’‘高’这三个属性，用0来表示反义的`
7.  `data[data==u'好']=1`
8.  `data[data==u'是']=1`
9.  `data[data==u'高']=1`
10. `data[data!=1]=0`
11. `x=data.iloc[:,:3].as_matrix().astype(int)`
12. `y=data.iloc[:,3].as_matrix().astype(int)`
13. 
14. `from keras.models import Sequential`
15. `from keras.layers import Dense`
16. 
17. `model=Sequential()#建模`
18. `model.add(Dense(units=10,activation='relu',input_dim=3))`
19. `model.add(Dense(units=1,activation='sigmoid'))#由于是0-1输出，用sigmoid函数作为激活函数`
20. `model.compile(loss='binary_crossentropy',optimizer='adam')`
21. `#编译模型，损失函数与优选算法的定义`
22. 
23. `#模型训练`
24. `model.fit(x,y,epochs=1000,batch_size=10)#训练模型，学习1000次，分为10组`
25. `yp=model.predict_classes(x,batch_size=5).reshape(len(y))#分类预测`
26. 
27. `from cm_plot import *`
28. `cm_plot(y,yp).show()`

(4)分类与测试评价

\-绝对误差与相对误差

![SouthEast 7][]\-平均绝对误差

![SouthEast 8][]

\-均方误差

![SouthEast 9][]

加入了平方，则避免了正负误差不能相加的情况

\-均方根误差

![SouthEast 10][]

代表了预测值的离散度，也称为标准误差

\-平均绝对百分误差

![SouthEast 11][]

\-Kappa统计

![SouthEast 12][]

\-识别准确度

![SouthEast 13][]

\-识别精确率

![SouthEast 14][]

\-反馈率

![SouthEast 15][]

\-ROC曲线

\-混淆矩阵

![70][]

识别数据集D包括T0个样本，每类模型分别包含Ti个数据（i=1,....,N）,采取某种识别算法构造分类器C,cmij表示i类模式被判断为j类模式的数据占i类模式样本总数的百分比

![70 1][]

![70 2][]

[SouthEast]: https://image.dandelioncloud.cn/pgy_files/images/2024/01/29/eaf7f4109fc74f698d889b2859ef40b6.png
[SouthEast 1]: https://image.dandelioncloud.cn/pgy_files/images/2024/01/29/670c0fe78406444aa32b0eb7dc545680.png
[SouthEast 2]: https://image.dandelioncloud.cn/pgy_files/images/2024/01/29/de4f6fad868f43c1969c8cb4685a2b7a.png
[SouthEast 3]: https://image.dandelioncloud.cn/pgy_files/images/2024/01/29/3559d277fc7a4da8adecd468a5416e3f.png
[SouthEast 4]: https://image.dandelioncloud.cn/pgy_files/images/2024/01/29/f0cc02a8b5554073b2ca1a3d44b95482.png
[SouthEast 5]: https://image.dandelioncloud.cn/pgy_files/images/2024/01/29/90912a5c9cca488892dce90445e3cb4c.png
[SouthEast 6]: https://image.dandelioncloud.cn/pgy_files/images/2024/01/29/3f7c1abbae1943609c9bd40156da501e.png
[SouthEast 7]: https://image.dandelioncloud.cn/pgy_files/images/2024/01/29/7dd7c478bf6544ed9f9f7320e1cbb3aa.png
[SouthEast 8]: https://image.dandelioncloud.cn/pgy_files/images/2024/01/29/1286cb7ace314443ac177e72d9704968.png
[SouthEast 9]: https://image.dandelioncloud.cn/pgy_files/images/2024/01/29/c0d32c0fd9bb4109beb2683cb2556b21.png
[SouthEast 10]: https://image.dandelioncloud.cn/pgy_files/images/2024/01/29/79b4761d4e814b29a7d107cdc3dc34c9.png
[SouthEast 11]: https://image.dandelioncloud.cn/pgy_files/images/2024/01/29/c6837eac643f4f55b38422e7c5da813c.png
[SouthEast 12]: https://image.dandelioncloud.cn/pgy_files/images/2024/01/29/ec8042bf4bd54b1eb481058fea47d667.png
[SouthEast 13]: https://image.dandelioncloud.cn/pgy_files/images/2024/01/29/704a4f58db544e6eb256fc2a940134b0.png
[SouthEast 14]: https://image.dandelioncloud.cn/pgy_files/images/2024/01/29/26ec39074e0b441f83e7522afc278397.png
[SouthEast 15]: https://image.dandelioncloud.cn/pgy_files/images/2024/01/29/c1a30033fbb94936867488bf88570c8c.png
[70]: https://image.dandelioncloud.cn/pgy_files/images/2024/01/29/c99b23ec0c3e4802b592d3f3f4f960ce.png
[70 1]: https://image.dandelioncloud.cn/pgy_files/images/2024/01/29/ea91d4fa3bb44553a9010c777124f951.png
[70 2]: https://image.dandelioncloud.cn/pgy_files/images/2024/01/29/1bc48061426343aa9e9cdee12971b6e0.png