我将使用所有功能并查看我的集群的分离性如何根据某些指标表现，例如，silhouette score

此外，在聚类之前缩放数据非常重要，因为 kmeans 是一种基于距离的算法。

heart_data = pd.read_csv("https://archive.ics.uci.edu/ml/machine-learning-databases/00519/heart_failure_clinical_records_dataset.csv")

from sklearn.cluster import KMeans
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler
from sklearn.metrics import silhouette_score

Features = heart_data.drop(["DEATH_EVENT"], axis = 1).columns

X = heart_data[Features]

sc = []
for i in range(2, 25):
    kmeans = Pipeline([("scaling",StandardScaler()),("clustering",KMeans(n_clusters=i, init='k-means++', max_iter=300, n_init=10, random_state=0))]).fit(X)
    score = silhouette_score(X, kmeans["clustering"].labels_)
    sc.append(score)
plt.plot(range(2, 25), sc, marker = "o")
plt.title('Silhouette')
plt.xlabel('Number of clusters')
plt.ylabel('Score')
plt.show()

您还可以尝试不同的功能组合，以使得分最高

出于可视化目的，您可以使用分解技术

from sklearn.decomposition import PCA
import matplotlib.pyplot as plt
plt.style.use("seaborn-whitegrid")

pca = Pipeline([("scaling",StandardScaler()),("decompositioning",PCA(n_components = 2))]).fit(X)

X2D = pca.transform(X)

plt.scatter(X2D[:,0],X2D[:,1], c = kmeans["clustering"].labels_, cmap = "RdYlBu")
plt.colorbar();

最后但并非最不重要的一点是，我建议在您的数据中使用诸如UMAP之类的流形投影，它可能会通过生成“定义明确的”集群来帮助您完成任务（可能但不一定，但值得尝试）

看，通过使用 UMAP，结果似乎有所改善：

代码：

# pip install umap-learn

heart_data = pd.read_csv("https://archive.ics.uci.edu/ml/machine-learning-databases/00519/heart_failure_clinical_records_dataset.csv")

from sklearn.cluster import KMeans, DBSCAN
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler
from sklearn.metrics import silhouette_score
from umap import UMAP

Features = heart_data.drop(["DEATH_EVENT"], axis = 1).columns

X = heart_data[Features]

sc = []
for i in range(2, 25):
    kmeans = Pipeline([("scaling",StandardScaler()),("umap",UMAP()),("clustering",KMeans(n_clusters=i, init='k-means++',random_state=0))]).fit(X)
    score = silhouette_score(X, kmeans["clustering"].labels_)
    sc.append(score)
plt.plot(range(2, 25), sc, marker = "o")
plt.title('Silhouette')
plt.xlabel('Number of clusters')
plt.ylabel('Score')
plt.show()

from sklearn.decomposition import PCA
import matplotlib.pyplot as plt
plt.style.use("seaborn-whitegrid")

kmeans = Pipeline([("scaling",StandardScaler()),("umap",UMAP()),("clustering",KMeans(n_clusters=3, init='k-means++',random_state=0))]).fit(X)

pca = Pipeline([("scaling",StandardScaler()),("umap",UMAP()),("decompositioning",PCA(n_components = 2))]).fit(X)

X2D = pca.transform(X)

plt.scatter(X2D[:,0],X2D[:,1], c = kmeans["clustering"].labels_, cmap = "RdYlBu")
plt.colorbar();

绘图显示 umap 投影的第一个和第二个主成分（它只是对所有特征在 2D 空间中的外观的投影）

颜色是集群 ID。即对于我们看到的每种颜色，算法（k-means）将每个观察值分配给哪个集群。