python实现ADASYN处理不平衡数据

ADASYN(Adaptive Synthetic Sampling)是一种用于解决不平衡数据问题的算法，它可以根据数据分布的密度进行自适应地合成少数类样本。下面是Python实现ADASYN处理不平衡数据的示例代码：

import numpy as np
from collections import Counter

def adasyn(X, y, n_neighbors=5, ratio=0.5, beta=1.0):
    """
    ADASYN算法处理不平衡数据
    :param X: 特征矩阵
    :param y: 标签向量
    :param n_neighbors: 每个少数类样本选择的近邻数
    :param ratio: 合成的少数类样本数目与原少数类样本数目之比
    :param beta: 分布密度偏向因子，控制合成样本在密度稀疏区域的生成量
    :return: 合成后的特征矩阵和标签向量
    """
    # 统计每个类别的样本数
    counter = Counter(y)
    majority_class = max(counter, key=counter.get)
    minority_class = min(counter, key=counter.get)
    n_samples = len(X)
    n_minority = counter[minority_class]
    n_synthetic = int(ratio * n_minority)

    # 计算每个样本的分布密度
    dist = np.zeros(n_samples)
    for i in range(n_samples):
        dist[i] = np.sum(np.square(X[i] - X), axis=1)
    dist /= np.max(dist)

    # 合成新的少数类样本
    synthetic_X = []
    synthetic_y = []
    for i in range(n_samples):
        if y[i] == minority_class:
            # 找到样本i的近邻
            neighbors = np.argsort(dist)[1:n_neighbors + 1]
            neighbors = neighbors[y[neighbors] == majority_class]
            if len(neighbors) > 0:
                # 根据密度比例计算合成样本的数量
                g = np.sum(dist[neighbors]) / len(neighbors)
                n = int(beta * g)
                for j in range(n):
                    # 生成合成样本
                    k = np.random.choice(neighbors)
                    diff = X[k] - X[i]
                    synthetic = X[i] + np.random.rand() * diff
                    synthetic_X.append(synthetic)
                    synthetic_y.append(minority_class)

    # 合并原始样本和合成样本
    synthetic_X = np.array(synthetic_X)
    synthetic_y = np.array(synthetic_y)
    X_resampled = np.vstack((X, synthetic_X))
    y_resampled = np.hstack((y, synthetic_y))

    return X_resampled, y_resampled

示例用法：

from sklearn.datasets import make_classification
from sklearn.model_selection import train_test_split
from sklearn.linear_model import LogisticRegression
from sklearn.metrics import classification_report

# 生成不平衡数据集
X, y = make_classification(n_samples=10000, n_features=20, n_informative=10, n_redundant=5,
                            weights=[0.9, 0.1], random_state=42)

# 划分训练集和测试集
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3, random_state=42)

# 使用ADASYN算法处理不平衡数据
X_resampled, y_resampled = adasyn(X_train, y_train, n_neighbors=5, ratio=0.5, beta=1.0)

# 训练模型
clf = LogisticRegression(random_state=42)
clf.fit(X_resampled, y_resampled)

# 在测试集上评估模型
y_pred = clf.predict(X_test)
print(classification_report(y_test, y_pred))

参考文献：[Haibo He, Yang Bai, Edwardo A. Garcia, and Shutao Li. ADASYN: Adaptive Synthetic Sampling for Imbalanced Learning. In: Proc. 2008 IEEE Intl. Joint Conf. on Neural Networks (IJCNN 2008), pp. 1322-1328, June 2008.](https://sci2s.ugr.es/keel/pdf/algorithm/congreso/2008-He-ijcnn.pdf)