Kaggle 比赛: 德国信用卡违约数据分析

数据描述

German Credit Data， 咱们来看看数据的格式,

A1 到 A15 为 15个不一样类别的特征，A16 为 label 列，一共有 690条数据，下面列举其中一条看成例子：

A1	A2	A3	A4	A5	A6	A7	A8	A9	A10	A11	A12	A13	A14	A15	A16
b	30.83	0	u	g	w	v	1.25	t	t	01	f	g	00202	0	+

Attribute Information:

A1:    b, a.
    A2:    continuous.
    A3:    continuous.
    A4:    u, y, l, t.
    A5:    g, p, gg.
    A6:    c, d, cc, i, j, k, m, r, q, w, x, e, aa, ff.
    A7:    v, h, bb, j, n, z, dd, ff, o.
    A8:    continuous.
    A9:    t, f.
    A10:    t, f.
    A11:    continuous.
    A12:    t, f.
    A13:    g, p, s.
    A14:    continuous.
    A15:    continuous.
    A16: +,-         (class attribute)

Missing Attribute Values:

37 cases (5%) have one or more missing values.  The missing
values from particular attributes are:

A1:  12
A2:  12
A4:   6
A5:   6
A6:   9
A7:   9
A14: 13

Class Distribution

+: 307 (44.5%)
-: 383 (55.5%)

数据处理与数据分析

下面展现一下数据处理流程，主要是处理了一下缺失值，而后根据特征按连续型和离散型进行分别处理，使用了 sklearn 里面的 LogisticRegression 包，下面的代码都有很详细的注释。

import pandas as pd
import numpy as np
import matplotlib as plt
import seaborn as sns

# 读取数据
data = pd.read_csv("./crx.data")

# 给数据增长列标签
data.columns = ["f1", "f2", "f3", "f4", "f5", "f6", "f7", "f8", "f9", "f10", "f11", "f12", "f13", "f14", "f15", "label"]

# 替换 label 映射
label_mapping = {
    "+": 1,
    "-": 0
}

data["label"] = data["label"].map(label_mapping)

# 处理缺省值的方法
data = data.replace("?", np.nan)

# 将 object 类型的列转换为 float型
data["f2"] = pd.to_numeric(data["f2"])
data["f14"] = pd.to_numeric(data["f14"])

# 连续型特征若是有缺失值的话，用它们的平均值替代
data["f2"] = data["f2"].fillna(data["f2"].mean())
data["f3"] = data["f3"].fillna(data["f3"].mean())
data["f8"] = data["f8"].fillna(data["f8"].mean())
data["f11"] = data["f11"].fillna(data["f11"].mean())
data["f14"] = data["f14"].fillna(data["f14"].mean())
data["f15"] = data["f15"].fillna(data["f15"].mean())

# 离散型特征若是有缺失值的话，用另一个不一样的值替代
data["f1"] = data["f1"].fillna("c")
data["f4"] = data["f4"].fillna("s")
data["f5"] = data["f5"].fillna("gp")
data["f6"] = data["f6"].fillna("hh")
data["f7"] = data["f7"].fillna("ee")
data["f13"] = data["f13"].fillna("ps")

tf_mapping = {
    "t": 1,
    "f": 0
}

data["f9"] = data["f9"].map(tf_mapping)
data["f10"] = data["f10"].map(tf_mapping)
data["f12"] = data["f12"].map(tf_mapping)

# 给离散的特征进行 one-hot 编码
data = pd.get_dummies(data)

from sklearn.linear_model import LogisticRegression

# 打乱顺序
shuffled_rows = np.random.permutation(data.index)

# 划分本地测试集和训练集
highest_train_row = int(data.shape[0] * 0.70)
train = data.iloc[0:highest_train_row]
loc_test = data.iloc[highest_train_row:]

# 去掉最后一列 label 以后的才是 feature
features = train.drop(["label"], axis = 1).columns

model = LogisticRegression()
X_train = train[features]
y_train = train["label"] == 1

model.fit(X_train, y_train)
X_test = loc_test[features]

test_prob = model.predict(X_test)
test_label = loc_test['label']

# 本地测试集上的准确率
accuracy_test = (test_prob == loc_test["label"]).mean()
print accuracy_test

0.835748792271

from sklearn import cross_validation, metrics

#验证集上的auc值
test_auc = metrics.roc_auc_score(test_label, test_prob)#验证集上的auc值
print test_auc

0.835748792271

简单使用了一下逻辑回归，发现准确率是 0.835748792271，AUC 值是 0.835748792271，效果还不错，接下来对模型进行优化来进一步提升准确率。