Kaggle 比賽: 德國信用卡違約數據分析

數據描述

German Credit Data， 咱們來看看數據的格式,

A1 到 A15 爲 15個不一樣類別的特徵，A16 爲 label 列，一共有 690條數據，下面列舉其中一條看成例子：

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%)

數據處理與數據分析

下面展現一下數據處理流程，主要是處理了一下缺失值，而後根據特徵按連續型和離散型進行分別處理，使用了 sklearn 裏面的 LogisticRegression 包，下面的代碼都有很詳細的註釋。

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

# 讀取數據
data = pd.read_csv("./crx.data")

# 給數據增長列標籤
data.columns = ["f1", "f2", "f3", "f4", "f5", "f6", "f7", "f8", "f9", "f10", "f11", "f12", "f13", "f14", "f15", "label"]

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

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

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

# 將 object 類型的列轉換爲 float型
data["f2"] = pd.to_numeric(data["f2"])
data["f14"] = pd.to_numeric(data["f14"])

# 連續型特徵若是有缺失值的話，用它們的平均值替代
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())

# 離散型特徵若是有缺失值的話，用另一個不一樣的值替代
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)

# 給離散的特徵進行 one-hot 編碼
data = pd.get_dummies(data)

from sklearn.linear_model import LogisticRegression

# 打亂順序
shuffled_rows = np.random.permutation(data.index)

# 劃分本地測試集和訓練集
highest_train_row = int(data.shape[0] * 0.70)
train = data.iloc[0:highest_train_row]
loc_test = data.iloc[highest_train_row:]

# 去掉最後一列 label 以後的纔是 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']

# 本地測試集上的準確率
accuracy_test = (test_prob == loc_test["label"]).mean()
print accuracy_test

0.835748792271

from sklearn import cross_validation, metrics

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

0.835748792271

簡單使用了一下邏輯迴歸，發現準確率是 0.835748792271，AUC 值是 0.835748792271，效果還不錯，接下來對模型進行優化來進一步提升準確率。