Kai Jia

Dr. Kai Jia


Research Focus & Metrics

Academic Career Length

4Years (2023-Present)

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

Representative Publications

View All Publications
22 March 2023 in IEEE Transactions on Emerging Topics in Computing
Verified
Journal Article
Closed access
KJKai Jia
Kai Jia
XYXiao Yu
Xiao Yu
CZChen Zhang
Chen Zhang
WHWenhua Hu
Wenhua Hu
DZDongdong Zhao
Dongdong Zhao
JXJianwen Xiang
Jianwen Xiang
Software aging, which is caused by the accumulation of errors in the system and the consumption of computing resources, tends to occur in long-running cloud service software systems. In practice, software aging prediction has proven to be useful in planning the time to trigger rejuvenation because it provides a prior estimate of future resource consumption. However, aging indicators (e.g., physical memory) in cloud may have the characteristics of long-term slow growth, medium-term seasonality variations (alternating peaks and troughs), and short-term irregular fluctuations. Unfortunately, most of the existing aging prediction methods (e.g., a statistical or single machine learning model) only focus on the accuracy of short-term prediction, while lacking the cognition of the medium and long-term variations of aging indicators and their functions in formulating the rejuvenation schedule (e.g., performing rejuvenation when the load is low can minimize interference to users). To address the above problems, this paper proposes a novel hybrid aging prediction framework to work on the prediction of memory resource consumption in cloud by applying a seasonal-trend decomposition procedure based on loess (STL) method and Gate Recurrent Unit (GRU) neural network, called the decomposition-based GRU (DGRU). The effectiveness of DGRU mainly has two aspects. One is the STL method as a preprocessing technology that can extract the trend, seasonality, and residual characteristics from memory utilization data. The other is that these characteristics can be well predicted separately by a well-designed GRU model, which can model the time-series relationship between the data. Experimental results show that our DGRU framework has superior performance compared with its competitors, including seven single models and six hybrid models. Our study illustrates that the DGRU is a promising solution for high-precision software aging prediction.

6 October 2025 in Sustainability
Verified
Journal Article
Open access
PDF file
ZWZhaohan Wang
Zhaohan Wang
KJKai Jia
Kai Jia
WZWenpeng Zhang
Wenpeng Zhang
CZChen Zhang
Chen Zhang
Particulate matter (PM) concentration, especially PM2.5, is a major culprit of environmental pollution from unreasonable energy system emissions that significantly affects visibility, climate, and public health. The prediction of PM2.5 concentration holds significant importance in the early warning and management of severe air pollution, since it enables the provision of guidance for scientific decision-making through the estimation of impending PM2.5 concentration. However, due to diversified human activities, seasonal factors and industrial emissions, the air quality data not only show local anomalous mutability, but also global dynamic change characteristics. This hinders existing PM2.5 prediction models from fully capturing the aforementioned characteristics, thereby deteriorating the model performance. To address these issues, this study proposes a framework integrating multi-scale temporal convolutional networks (TCNs) and a transformer network (called MSTTNet) for PM2.5 concentration prediction. Specifically, MSTTNet uses multi-scale TCNs to capture the local correlations of meteorological and pollutant data in a fine-grained manner, while using transformers to capture the global temporal relationships. The proposed MSTTNet’s performance has been validated on various air quality benchmark datasets in the cities of China, including Beijing, Shanghai, Chengdu, and Guangzhou, by comparing to its eight compared models. Comprehensive experiments confirm that the MSTTNet model can improve the prediction performance of 2.42%, 2.17%, 2.87%, and 0.34%, respectively, with respect to four evaluation indicators (i.e., Mean Absolute Error, Root Mean Square Error, Mean Absolute Percentage Error, and R-square), relative to the optimal baseline model. These results confirm MSTTNet’s effectiveness in improving the accuracy of PM2.5 concentration prediction.

22 July 2024 in Software Quality Journal
Verified
Journal Article
Closed access
PDF file
KJKai Jia
Kai Jia
XYXiao Yu
Xiao Yu
CZChen Zhang
Chen Zhang
WXWenzhi Xie
Wenzhi Xie
DZDongdong Zhao
Dongdong Zhao
JXJianwen Xiang
Jianwen Xiang

10 September 2024 in Electronics
Verified
Journal Article
Open access
PDF file
KJKai Jia
Kai Jia
JXJun Xiang
Jun Xiang
BLBaoxia Li
Baoxia Li
Cloud infrastructures are designed to provide highly scalable, pay-as-per-use services to meet the performance requirements of users. The workload prediction of the cloud plays a crucial role in proactive auto-scaling and the dynamic management of resources to move toward fine-grained load balancing and job scheduling due to its ability to estimate upcoming workloads. However, due to users’ diverse usage demands, the changing characteristics of workloads have become more and more complex, including not only short-term irregular fluctuation characteristics but also long-term dynamic variations. This prevents existing workload-prediction methods from fully capturing the above characteristics, leading to degradation of prediction accuracy. To deal with the above problems, this paper proposes a framework based on a dual-channel temporal convolutional network and transformer (referred to as DuCFF) to perform workload prediction. Firstly, DuCFF introduces data preprocessing technology to decouple different components implied by workload data and combine the original workload to form new model inputs. Then, in a parallel manner, DuCFF adopts the temporal convolution network (TCN) channel to capture local irregular fluctuations in workload time series and the transformer channel to capture long-term dynamic variations. Finally, the features extracted from the above two channels are further fused, and workload prediction is achieved. The performance of the proposed DuCFF’s was verified on various workload benchmark datasets (i.e., ClarkNet and Google) and compared to its nine competitors. Experimental results show that the proposed DuCFF can achieve average performance improvements of 65.2%, 70%, 64.37%, and 15%, respectively, in terms of Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE) and R-squared (R2) compared to the baseline model CNN-LSTM.

31 July 2024 in Journal of Systems and Software
Verified
Journal Article
Closed access
KJKai Jia
Kai Jia
XYXiao Yu
Xiao Yu
CZChen Zhang
Chen Zhang
WXWenzhi Xie
Wenzhi Xie
DZDongdong Zhao
Dongdong Zhao
JXJianwen Xiang
Jianwen Xiang

15 May 2023 in Research Square
Verified
Preprint
Open access
PDF file
CZChen Zhang
Chen Zhang
CDCotroneo Domenico
Cotroneo Domenico
PRPietrantuono Roberto
Pietrantuono Roberto
Roberto NatellaRoberto Natella
Roberto Natella
XYXiao Yu
Xiao Yu
JXJianwen Xiang
Jianwen Xiang
Software aging refers to the accumulation of error conditions over time in long-running software systems, which can lead to decreased performance and an increased likelihood of failures. Aging-Related Bug Prediction (ARBP) was introduced to predict the Aging-related Bugs (ARBs) hidden in the systems by using features extracted from source code. ARBs include memory leaks, storage problems, unreleased files, socket exceptions, unreleased file handles, disk fragmentation and so on. Previous research in Software Defect Prediction (SDP) indicated that using feature selection techniques to select a subset of representative features to use could enhance the performance of classification models. However, considering the difference between ARB features and SDP features, blindly applying the method performed well in SDP to pre-process the ARBs dataset may not necessarily improve the performance of the ARBP model, and could potentially result in a decline in performance. To address this limitation, 22 feature selection methods with 21 classifiers embedded in the most used ARBP model on four benchmark datasets from real-world software projects, and six different evaluation indicators were employed to assess the performance of ARBP models comprehensively. Our experiment results showed that: (1) The filter-based feature ranking method called SVMF performed the best on the ARBP, and the filter-based feature subset selection method ConBF performs the worst on the ARBP task. (2) Using the statistic-based classifiers as the base classification model embedded with the SVMF can perform the best, the Naive Bayes classifier always achieves the best performance. Researchers are recommended to first consider CountLineBlank, CountLineComment, and MaxCyclomaticModified features for the ARBP task.(3) The feature selection method ConBF, which performed the best in conventional SDP was not optimal for our specific task. This highlights the unique nature of aging-related features and underscores the need for a tailored feature selection method. Based on these findings, we recommend using SVMF with the Naive Bayes classifier when building ARBP models, in our study, this combination can improve the Balance performance by 18\% and Recall by 25.9% compared with no feature selection for ARBP.