In [1]:
import pandas as pd
pd.set_option('display.max_columns', 50)
import numpy as np
import os
os.chdir('D:\Data\Projects\Klassifikation\Heart Disease')
import plotly_express as px
import matplotlib.pyplot as plt
plt.style.use('Solarize_Light2')
plt.rcParams['font.size'] = 15
from IPython.core.pylabtools import figsize
figsize(10, 10)
import seaborn as sns
from warnings import filterwarnings
filterwarnings('ignore')
In [2]:
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler
from sklearn.metrics import roc_curve, classification_report, accuracy_score
from sklearn.metrics import f1_score, confusion_matrix, roc_auc_score
from sklearn.model_selection import GridSearchCV
from sklearn import model_selection
from sklearn.svm import SVC
In [3]:
df = pd.read_csv('heart.csv')
df.shape
Out[3]:
In [4]:
df.head(5)
Out[4]:
In [5]:
x = df.drop('target', axis=1)
y = df.target
x.shape, y.shape
Out[5]:
In [6]:
X_train, X_test, y_train, y_test = train_test_split(x, y, test_size=0.2, random_state=42)
X_train.shape, X_test.shape, y_train.shape, y_test.shape
Out[6]:
Standardisieren der Daten mit Standard Scaler¶
Mittelwert bei 0, Standardabweichung 1
In [11]:
sc = StandardScaler()
sc.fit(X_train)
X_train =sc.transform(X_train);
X_test =sc.transform(X_test);
Baseline Support Vector Machines Classifier¶
In [74]:
svc = SVC(probability=True)
svc.fit(X_train, y_train)
y_pred = svc.predict(X_test)
y_pred_proba = svc.predict_proba(X_test)
In [75]:
accuracy_score(y_pred, y_test)
Out[75]:
GridSearch für beste Hyperparameter¶
- C Penalty Parameter of the error term. Ein kleiner Wert macht eine Hyperplane mit schmalem Margin.
- kernel
linear
polynomial
radial basis function: kann mit Daten umgehen, die nicht linear getrennt werden können
sigmoid - gamma Ein kleinerer Wert passt sich lose an, ein höherer Wert von gamma passt sich genau an
In [12]:
svc.get_params()
Out[12]:
In [18]:
parameters = {'C': [1,10,100,1000],
'gamma': [0.1, 0.001, 0.0001, 0.00001],
'kernel': ['rbf', 'poly']}
In [19]:
grid_search = GridSearchCV(svc, parameters, cv=10, n_jobs=-1, scoring= 'accuracy', verbose=1)
grid_search.fit(X_train, y_train)
print(grid_search.best_score_)
print(grid_search.best_params_)
Das Model mit den besten Parametern testen
In [20]:
y_pred_grid_search = grid_search.predict(X_test)
In [21]:
accuracy_score(y_pred_grid_search, y_test)
Out[21]:
Polynomial Features und Interaction Features¶
In [22]:
df_poly = pd.read_csv('df_poly.csv')
df_poly.head()
Out[22]:
In [23]:
x_p = df_poly.drop('target', axis=1)
x_p.shape, y.shape
Out[23]:
In [35]:
X_train_p, X_test_p, y_train, y_test = train_test_split(x_p, y, test_size=0.2, random_state=4)
X_train_p.shape, X_test_p.shape, y_train.shape, y_test.shape
Out[35]:
In [36]:
sc.fit(X_train_p)
X_train_p =sc.transform(X_train_p);
X_test_p =sc.transform(X_test_p);
Mit default Parametern¶
In [37]:
svc_p = SVC()
In [38]:
svc_p.fit(X_train_p, y_train)
Out[38]:
In [39]:
y_p_pred = svc_p.predict(X_test_p)
In [40]:
accuracy_score(y_test, y_p_pred)
Out[40]:
Mit besten Parametern¶
In [43]:
svc_p_ = SVC(C= 100, gamma= 0.0001, kernel= 'rbf', probability=True)
svc_p_.fit(X_train_p, y_train)
y_p_pred_ = svc_p_.predict(X_test_p)
accuracy_score(y_test, y_p_pred_)
Out[43]:
In [78]:
print(classification_report(y_test, y_p_pred))
Die Kombination aus besten Hyperparametern und Feature Engineering konnte das Ergebnis von 54% Accuracy auf 90% verbessern!¶
In [44]:
y_p_proba = svc_p_.predict_proba(X_test_p)
ROC Curves der vier Models¶
In [46]:
# Beste Hyperparameter, Polynomial Datensatz
fpr, tpr, thresholds = roc_curve(y_test, y_p_proba[:,1], drop_intermediate=False)
In [49]:
fpr0, tpr0, thresholds0 = roc_curve(y_test, y_pred_proba[:,1], drop_intermediate=False)
In [60]:
figsize(10, 7)
plt.rcParams['font.size'] = 20
plt.plot(fpr, tpr, label = 'Best Params')
plt.plot(fpr0, tpr0, label = 'Baseline')
plt.plot(np.linspace(0,1,10), np.linspace(0,1,10), '--g', label="Random")
plt.xlim([-0.05, 1.05])
plt.ylim([-0.05, 1.05])
plt.title('ROC Curves Heart Disease')
plt.xlabel('False Positive Rate', fontsize=18)
plt.ylabel('True Positive Rate', fontsize=18)
plt.grid(True)
plt.legend(loc="lower right")
Out[60]:
Support Vectors¶
In [62]:
svc_p_.support_vectors_
Out[62]:
Plotten geht nur mit kernel = linear