深度學習在不平衡數據集之研究

Abstract

本文旨在回顧利用深度學習處理不平衡數據集和異常偵測的方法，我們 從 MNIST 生成兩個高度不平衡數據集，不平衡比率高達 2500 並應用在多 元分類任務跟二元分類任務上，在二元分類任務中第 0 類為少數類；而在 多元分類任務中少數類為第 0、1、4、6、7 類，我們利用卷積神機網路來 訓練我們的模型。在異常偵測方面，我們用預先訓練好的手寫辨識 CNN 模 型來判斷其他 18 張貓狗的圖片是否為手寫辨識圖片。\n由於數據的高度不平衡，原始分類模型的表現不盡理想。因此，在不同 的分類任務上，我們分別利用 6 個和 7 個不同的方法來調整我們的模型。我 們發現新的損失函數 Focalloss 在多元分類任務表現最好，而在二元分類任\n務中隨機過採樣的表現最佳，但是成本敏感學習的方法並不適用於我們所\n生成的不平衡數據集。我們利用信心估計讓分類器成功判斷所有貓狗圖片\n皆不是手寫辨識圖片。

This paper is a survey on deep learning with imbalanced data sets and anomaly detection. We create two imbalanced data sets from MNIST for multi­-classification task with minority classes 0,1,4,6,7 and binary classification task with minority class 0. Our data sets are highly imbalanced with imbalanced rate ρ = 2500 and we use convolutional neural network(CNN) for training. In anomaly detection,we use the pretrained CNN handwriting classifier to decide the 18 cat and dog pictures are handwriting pictures or not.\nDue to the data set is imbalanced, the baseline model have poor performance on minority classes. Hence, we use 6 and 7 different methods to adjust our model. We find that the focal loss function and random over­-sampling(ROS) have best performance on multi­-classification task and binary classification task on our imbalanced data sets but the cost sensitive learning method is not suitable for our imbalanced data sets. By confidence estimation, our classifier successfully judge all the pictures of cat and dog are not handwriting picture.