Chang Zhao

Ms. Chang Zhao

PLA Information Engineering University, Zhengzhou, Henan, China
orcid

Research Focus & Metrics

Academic Career Length

7Years (2020-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
21 April 2022 in Electronics
Verified
Journal Article
Open access
PDF file
QLQing Li
Qing Li
CZChang Zhao
Chang Zhao
XHXintai He
Xintai He
KCKun Chen
Kun Chen
RWRunze Wang
Runze Wang
The imbalance of network data seriously affects the classification performance of algorithms. Most studies have only used a rough description of data imbalance with less exploration of the specific factors affecting classification performance, which has resulted in difficulty putting forward targeted solutions. In this paper, we find that the impact of medium categories on classification performance cannot be ignored, and therefore propose the concept of partial balance, consisting of Class Number of Partial Balance (β) and Balance Degree of Partial Samples (μ). Combined with Global Slope (α), a parameterized model is established to describe the difference of imbalanced datasets. Experiments are performed on the Moore Dataset and CICIDS 2017 Dataset. The experiment’s results on Random Forest, Decision Tree and Deep Neural Network show increasing α is a conducive step in the performance improvement of minority classes and overall classes. When β of dominant categories increases, that of inferior classes decreases, which results in a decrease in the average performance of minority classes. The lower μ is, the closer the sample size of medium classes is to the minority classes, and the better the average performance is. Based on the conclusions, we propose and verify some basic strategies by various classical algorithms.

1 March 2021 in IEEE Access
Verified
Journal Article
Open access
QLQing Li
Qing Li
YJYonghui Ju
Yonghui Ju
CZChang Zhao
Chang Zhao
XHXintai He
Xintai He
Encrypted network data classification has received considerable attention in the industry and research communities for a long time. However, the emergence of new private applications and encryption protocols has brought more new challenges. The primary task of classification is to determine whether the sample is encrypted. When the specification of private protocol is unpublished, only the whole sample can be processed, and thus the unencrypted field which coexists with the encrypted field will seriously lower the classification effect. To tackle this problem, an algorithm based on data reconstruction and moment eigenvector is proposed, which can not only estimate the encryption result but also locate the encrypted field in each data sample. In the algorithm, the encryption probability sequence is calculated firstly by data reconstruction and CNN (Convolutional Neural Network) model transfer. Then, based on the derivative of the encryption probability sequence, the suspected encrypted field set is generated. Finally, encrypted field matching is performed based on the similarity of the four-dimensional moment eigenvector, and thus the possible encrypted field of each sample is obtained. The proposed algorithm achieved a recall rate of 93% and a precision rate of 72% in an experiment of distinguishing the encrypted/unencrypted ones of complex data. The forward coverage, reverse coverage, and F1 value in identifying encrypted field reached 89%, 90%, and 90%, respectively. Compared with the encrypted field matching methods based on the K-Nearest Neighbor algorithm, Dynamic Time Warping, Runs test, and Frequency test, the method proposed in this paper exhibited salient advantages.

28 October 2020
Verified
Conference Paper
Closed access
YJYonghui Ju
Yonghui Ju
QLQing Li
Qing Li
CZChang Zhao
Chang Zhao
To solve the problem of classification for the sparsely labeled encrypted Discrete Sequence Protocol Data (DSPD), a two-stage semi-supervised classification method based on Generative Adversarial Networks (GAN) is proposed. As data annotation for DSPD is difficult, the semi-supervised GAN is used for classification by mining the data distribution hidden in unlabeled data. Meanwhile, a data filtering method based on traditional GAN is put forward to avoid the adverse effects caused by the irrelevant data. In addition, to directly process the raw data, the discriminator in the semi-supervised GAN is constructed by a Long Short-Term Memory (LSTM) network. Experiments results show that when the labeling rate in the class-balanced mixed data is as low as 0.48%, the average accuracy and F1 can reach 96%. Compared with the supervised classification based on LSTM, the accuracy and F1 of the proposed method are improved by more than 10 percentage points on average. For the class-imbalanced data, when the proportion of smallest class is not less than 10%, the accuracy and F1 can be maintained above 90%.