数据挖掘实战

导入数据，将 标签（Y）和样本（X) 分开来

import pandas as pd
X = pd.read_csv("data.csv")
y = X["status"]
X.drop(["status"],axis=1)
X.head(5)

将数据分成训练集和测试集

from sklearn.model_selection import train_test_split
X_train, X_test, Y_train, Y_test = train_test_split(X, y, test_size=0.30, random_state=2018)

查看数据的信息

X_train.info()

<class 'pandas.core.frame.DataFrame'>
Int64Index: 3327 entries, 110 to 1274
Data columns (total 90 columns):
Unnamed: 0                                    3327 non-null int64
custid                                        3327 non-null int64
trade_no                                      3327 non-null object
bank_card_no                                  3327 non-null object
low_volume_percent                            3325 non-null float64
middle_volume_percent                         3325 non-null float64
take_amount_in_later_12_month_highest         3327 non-null int64
trans_amount_increase_rate_lately             3325 non-null float64
trans_activity_month                          3325 non-null float64
trans_activity_day                            3325 non-null float64
transd_mcc                                    3325 non-null float64
trans_days_interval_filter                    3321 non-null float64
trans_days_interval                           3325 non-null float64
regional_mobility                             3325 non-null float64
student_feature                               1208 non-null float64
repayment_capability                          3327 non-null int64
is_high_user                                  3327 non-null int64
number_of_trans_from_2011                     3325 non-null float64
first_transaction_time                        3325 non-null float64
historical_trans_amount                       3327 non-null int64
historical_trans_day                          3325 non-null float64
rank_trad_1_month                             3325 non-null float64
trans_amount_3_month                          3327 non-null int64
avg_consume_less_12_valid_month               3325 non-null float64
abs                                           3327 non-null int64
top_trans_count_last_1_month                  3325 non-null float64
avg_price_last_12_month                       3327 non-null int64
avg_price_top_last_12_valid_month             3249 non-null float64
reg_preference_for_trad                       3325 non-null object
trans_top_time_last_1_month                   3321 non-null float64
trans_top_time_last_6_month                   3321 non-null float64
consume_top_time_last_1_month                 3321 non-null float64
consume_top_time_last_6_month                 3321 non-null float64
cross_consume_count_last_1_month              3037 non-null float64
trans_fail_top_count_enum_last_1_month        3314 non-null float64
trans_fail_top_count_enum_last_6_month        3314 non-null float64
trans_fail_top_count_enum_last_12_month       3314 non-null float64
consume_mini_time_last_1_month                3312 non-null float64
max_cumulative_consume_later_1_month          3327 non-null int64
max_consume_count_later_6_month               3321 non-null float64
railway_consume_count_last_12_month           3319 non-null float64
pawns_auctions_trusts_consume_last_1_month    3327 non-null int64
pawns_auctions_trusts_consume_last_6_month    3327 non-null int64
jewelry_consume_count_last_6_month            3319 non-null float64
status                                        3327 non-null int64
source                                        3327 non-null object
first_transaction_day                         3325 non-null float64
trans_day_last_12_month                       3325 non-null float64
id_name                                       3125 non-null object
apply_score                                   3107 non-null float64
apply_credibility                             3107 non-null float64
query_org_count                               3107 non-null float64
query_finance_count                           3107 non-null float64
query_cash_count                              3107 non-null float64
query_sum_count                               3107 non-null float64
latest_query_time                             3107 non-null object
latest_one_month_apply                        3107 non-null float64
latest_three_month_apply                      3107 non-null float64
latest_six_month_apply                        3107 non-null float64
loans_score                                   3112 non-null float64
loans_credibility_behavior                    3112 non-null float64
loans_count                                   3112 non-null float64
loans_settle_count                            3112 non-null float64
loans_overdue_count                           3112 non-null float64
loans_org_count_behavior                      3112 non-null float64
consfin_org_count_behavior                    3112 non-null float64
loans_cash_count                              3112 non-null float64
latest_one_month_loan                         3112 non-null float64
latest_three_month_loan                       3112 non-null float64
latest_six_month_loan                         3112 non-null float64
history_suc_fee                               3112 non-null float64
history_fail_fee                              3112 non-null float64
latest_one_month_suc                          3112 non-null float64
latest_one_month_fail                         3112 non-null float64
loans_long_time                               3112 non-null float64
loans_latest_time                             3112 non-null object
loans_credit_limit                            3112 non-null float64
loans_credibility_limit                       3112 non-null float64
loans_org_count_current                       3112 non-null float64
loans_product_count                           3112 non-null float64
loans_max_limit                               3112 non-null float64
loans_avg_limit                               3112 non-null float64
consfin_credit_limit                          3112 non-null float64
consfin_credibility                           3112 non-null float64
consfin_org_count_current                     3112 non-null float64
consfin_product_count                         3112 non-null float64
consfin_max_limit                             3112 non-null float64
consfin_avg_limit                             3112 non-null float64
latest_query_day                              3107 non-null float64
loans_latest_day                              3112 non-null float64
dtypes: float64(70), int64(13), object(7)
memory usage: 2.3+ MB

将一些没用的特征删除

##将只有一个值的列删除
X_train.drop(["bank_card_no"],axis=1,inplace=True)
##删除身份证号
X_train.drop(["trade_no"],axis=1,inplace=True)
##删除名字
X_train.drop(['id_name'],axis=1,inplace=True)
##删除custid
X_train.drop(['custid'],axis=1,inplace=True)
X_train.drop(['loans_latest_time', 'latest_query_time'], axis=1,inplace=True)
X_train.drop(['source'], axis=1,inplace=True)

把值转为离散型数据

X_train['reg_preference_for_trad'].value_counts()
//
一线城市    2380
三线城市     747
境外       103
二线城市      92
其他城市       3
//
def get_change(dt):
    if dt == '一线城市':
        return 1
    if dt == '二线城市':
        return 2
    if dt == '三线城市':
        return 3
    if dt == '境外':
        return 4
    else :
        return 5

X_train['reg_preference_for_trad'] = np.array([get_change(x) for x in X_train['reg_preference_for_trad']])
X_train['reg_preference_for_trad'].value_counts()
//
1    2380
3     747
4     103
2      92
5       5
Name: reg_preference_for_trad, dtype: int64

处理缺失值，将缺失值填充为那一列的均值

cols = X_train.columns
for col in cols:
    X_train[col] = X_train[col].fillna(X_train[col].mean())

#方差选择法

from sklearn.feature_selection import VarianceThreshold
VarianceThreshold(threshold=3).fit_transform(X_train）

#L1方法

lsvc = LinearSVC(C=0.01, penalty="l1", dual=False).fit(X_train，y_train)
model = SelectFromModel(lsvc, prefit=True)
X_new_1 = model.transform(X_train)
X_new_1.shape  #特征减少了

#决策树方法

clf = ExtraTreesClassifier().fit(X_train，y_train)
clf.feature_importances_  
model = SelectFromModel(clf, prefit=True)
X_new_2 = model.transform(X_train)
X_new_2.shape #特征减少了

#随即森林特征选择 参考别人的

from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import StratifiedKFold

forest = RandomForestClassifier(n_estimators=100, random_state=0,n_jobs=1)
forest.fit(X_train，y_train)
importance = forest.feature_importances_
imp_result = np.argsort(importance)[::-1]
columns = []
for i in range(x_train.shape[1]):
    print("%2d. %-*s %f"%(i+1, 30, names[i], importance[imp_result[i]]))
    columns.append(names[i])
columns = columns[0:44]    
data =x_train [columns]
 
def RF_classify(train_data, train_label, test_data, test_label):
    forest = RandomForestClassifier(n_estimators=15, random_state=0, n_jobs=-1)  
    forest.fit(x_train, y_train) #training

    ac_train = forest.score(x_test，y_test)
    return ac_train
 
skf = StratifiedKFold(n_splits=10)#10折交叉
ac = []
for train_index, test_index in skf.split(data1,label):
    x_train, x_test = data1.iloc[train_index,:], data1.iloc[test_index,:]
    y_train, y_test = label.iloc[train_index],label.iloc[test_index]
    acc = RF_classify(x_train, y_train, x_test, y_test)
  #acc = lightgbm_classify(x_train, y_train, x_test, y_test)
    ac.append(acc)

开始调模型
//逻辑回归

from sklearn.metrics import accuracy_score
from sklearn.linear_model import LogisticRegression
ltc = LogisticRegression()
ltc.fit(X_train, y_train)
pre = ltc.predict(X_test)
accuracy_score(y_test, pre)

//SVM

from sklearn.svm import SVC
svc = SVC()
svc.fit(X_train, y_train)
pre = svc.predict(X_test)
accuracy_score(y_test, pre)

//RF

from sklearn.ensemble import RandomForestClassifier
RF = RandomForestClassifier()
RF.fit(X_train, y_train)
pre  = RF.predict(X_test)
accuracy_score(y_test, pre )

//XGboost

from xgboost import XGBClassifier
xgb = XGBClassifier()
xgb.fit(X_train, y_train)
pre = xgb.predict(X_test)
accuracy_score(pre, y_test)

预测结果都差不多 但xgboost和RF的效果会好一点，没有进行调参。