在實務上，資料很有可能不是完美的，特徵缺失就是其中一類問題。比方說，有某個預測健康狀況的模型，需要測量四肢的血壓，但是可能有人剛好骨折打石膏，因此其中幾肢的血壓無法測量。而依據資料中缺值的欄位或比率分布等狀況，則各自有不同的方法適合來處理，包含但不限於：

If you apply the example above to a real competition or research task, you may want to first save the models trained during cross validation, then load the model with better validation performance from another file and use it to predict the test set. This avoids wasting time using a poorly performing model to make predictions. If you need to save and load models, you can refer to Python's built-in pickle library.

上述的「很低」、「幾乎」等形容詞，可能需要根據資料狀況等方面，來決定較具體的標準。

The adjectives above, such as "very low" and "almost," may need more specific criteria depending on the actual data situation.

以下範例，會示範當隨機缺值比率從 1~10% 時，補特定值與使用 KNN Imputer 的模型辨識效果差異：

The example below demonstrates the difference in model recognition performance between filling in a fixed value and using KNN Imputer, when the random missing rate ranges from 1% to 10%:

import matplotlib.pyplot as plt
import numpy as np
from sklearn.datasets import load_wine
from sklearn.impute import KNNImputer
from sklearn.linear_model import LogisticRegression
from sklearn.metrics import accuracy_score
from sklearn.model_selection import train_test_split

np.random.seed(0)

dataset = load_wine()
X_train, X_test, y_train, y_test = train_test_split(
	dataset.data, dataset.target, test_size=0.2, random_state=42
)
print('Training data shapes:', X_train.shape, y_train.shape)
print('Test data shapes:', X_test.shape, y_test.shape)

acc_fill = []
acc_knn = []
missing_rates = np.arange(0.01, 0.11, 0.01)
for rate in missing_rates:

	X_train_miss = X_train.copy()
	X_test_miss = X_test.copy()

	mask_train = np.random.rand(*X_train_miss.shape) < rate
	mask_test = np.random.rand(*X_test_miss.shape) < rate

	X_train_miss[mask_train] = np.nan
	X_test_miss[mask_test] = np.nan

	# Fill by fix val
	X_train_fill = np.nan_to_num(X_train_miss, nan=0)
	X_test_fill = np.nan_to_num(X_test_miss, nan=0)

	# Fill by mean
	# col_mean = np.nanmean(X_train_miss, axis=0)
	# X_train_fill = np.where(np.isnan(X_train_miss), col_mean, X_train_miss)
	# X_test_fill = np.where(np.isnan(X_test_miss), col_mean, X_test_miss)

	model = LogisticRegression(max_iter=500)
	model.fit(X_train_fill, y_train)
	pred = model.predict(X_test_fill)

	acc_fill.append(accuracy_score(y_test, pred) * 100)

	# Impute by KNN
	imputer = KNNImputer(n_neighbors=5)

	X_train_knn = imputer.fit_transform(X_train_miss)
	X_test_knn = imputer.transform(X_test_miss)

	model = LogisticRegression(max_iter=500)
	model.fit(X_train_knn, y_train)
	pred = model.predict(X_test_knn)

	acc_knn.append(accuracy_score(y_test, pred) * 100)

plt.plot(missing_rates * 100, acc_fill, '.-', label='Impute by fix val')
plt.plot(missing_rates * 100, acc_knn, '.-', label='Impute by KNN')
plt.xlabel("Missing rate (%)")
plt.ylabel("Accuracy (%)")
plt.legend()

plt.show()

在上述範例中：

In the example above: