本文介紹如何利用機器學習模型根據各種特徵預測葡萄酒質量。從這裡下載分析數據集。
葡萄酒數據集包含以下特徵：
Input variables (based on physicochemical tests):
fixed acidity, volatile acidity, citric acid, residual sugar,
chlorides,free sulfur dioxide,total sulfur dioxide,density,
pH,sulphates, alcohol
Output variables:
quality (score between 0 and 10)
首先通過導入所需的Python庫並加載白葡萄酒和紅葡萄酒的csv文件來加載兩個數據集。
#import the libraries
import pandas as pd
import numpy as np
import seaborn as sns
import matplotlib.pyplot as plt
# load the files
df_red = pd.read_csv(“winequality-red.csv”, sep=”;”)
df_white = pd.read_csv(“winequality-white.csv”, sep=”;”)
將這兩個dataframes 合併起來分析。Python代碼如下：
df = pd.concat([df_red, df_white], axis=0)
檢查是否有任何空列
df.isnull().sum()
fixed acidity 0
volatile acidity 0
citric acid 0
residual sugar 0
chlorides 0
free sulfur dioxide 0
total sulfur dioxide 0
density 0
pH 0
sulphates 0
alcohol 0
quality 0
找出輸出(質量)變量與所有輸入變量之間的相關性，Python實現如下：
# identify the correlation
plt.subplots(figsize=(20,15))
corr = df.corr()
sns.heatmap(corr,square=True, annot=True)
一些如酒精，檸檬酸，遊離二氧化硫，pH值呈正相關，質量會有所改善，而密度，殘糖和酸度會對質量產生負面影響。
讓我們確定前6個相關特徵。Python代碼如下：
# pick the top 6 highly correlating columns
cols = corr.nlargest(6, ‘quality’)[‘quality’].index
corrcoef = np.corrcoef(df[cols].values.T)
# correlation plotted against the top columns
plt.subplots(figsize=(20,15))
corr = df.corr()
sns.heatmap(corrcoef,square=True, annot=True, xticklabels= cols.values, yticklabels=cols.values)
通過繪製直方圖來分析數據的分佈
使用機器學習中的sklearn庫，將數據集拆分為測試和訓練數據集，我使用了20％的數據作為測試數據集。Python代碼如下：
y = df[“quality”]
X = df.drop(“quality”, axis=1)
from sklearn.model_selection import train_test_split
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=0)
由於不同的列具有不同的值，因此您需要歸一化值以獲得準確的預測結果。我在這裡使用StandardScaler庫。您也可以使用MinMaxScaler方法。
from sklearn.preprocessing import StandardScaler
scaler = StandardScaler()
X_train = scaler.fit_transform(X_train)
X_test = scaler.fit_transform(X_test)
現在，我將根據各種算法擬合我的訓練數據，並根據測試值確定預測輸出的準確性。Python實現如下：
from sklearn.metrics import accuracy_score, confusion_matrix
from sklearn.linear_model import LogisticRegression
logreg = LogisticRegression()
logreg.fit(X_train, y_train)
pred_logreg = logreg.predict(X_test)
accuracy = accuracy_score(pred_logreg, y_test)
print("Logreg Accuracy Score %.2f" % accuracy)
cm = confusion_matrix(pred_logreg, y_test)
knn = KNeighborsClassifier(n_neighbors=1)
knn.fit(X_train, y_train)
pred_knn = knn.predict(X_test)
accuracy = accuracy_score(pred_knn, y_test)
print("Knn Accuracy Score %.2f" % accuracy)
from sklearn.svm import SVC
svc = SVC()
svc.fit(X_train, y_train)
pred_svc =svc.predict(X_test)
accuracy = accuracy_score(pred_svc, y_test)
print("SVC Accuracy Score %.2f" % accuracy)
dtree = DecisionTreeClassifier()
dtree.fit(X_train, y_train)
pred_tree =dtree.predict(X_test)
accuracy = accuracy_score(pred_tree, y_test)
print("DTree Accuracy Score %.2f" % accuracy)
from sklearn.ensemble import RandomForestClassifier
rf = RandomForestClassifier()
rf.fit(X_train, y_train)
pred_rf =rf.predict(X_test)
accuracy = accuracy_score(pred_rf, y_test)
print("Random Forest Accuracy Score %.2f" % accuracy)
我嘗試了各種算法，包括Logistic迴歸，決策樹，隨機森林，KNN和SVC。
隨機森林為我提供更好的準確性（64％）
Logreg Accuracy Score 0.53
Knn Accuracy Score 0.62
SVC Accuracy Score 0.57
DTree Accuracy Score 0.55
Random Forest Accuracy Score 0.64
將前10條記錄的測試數據與預測數據進行比較，結果表明，其中有2條記錄的質量預測與測試結果不同
閱讀更多 不靠譜的貓 的文章
