ISWC 2024 ADDKG

该数据集包含了五天的标注数据，其中三天用于训练，两天用于测试。由于异常数据仅占总数据的约4%，这给数据集带来了不平衡的挑战。为了在模型训练过程中缓解这种不平衡，我们对损失函数应用了类别权重。总的来说，这个数据集在规模上涵盖了五天的数据，异常类别存在显著的不平衡，其任务是对二分类异常进行检测。This dataset comprises five days of annotated data, with three days reserved for training and the remaining two days for testing. As anomalous data accounts for only approximately 4% of the total dataset, it poses a significant class imbalance challenge. To mitigate this imbalance during model training, we applied class weights to the loss function. Overall, this dataset spans five days of data, exhibits a notable imbalance in the anomalous class, and its task is binary anomaly detection.