缺失数据填充是数据分析处理领域的一个重要研究课题。特别是在采集数据量较少的情况下，缺失数据填充的难度极大。针对这个问题，提出一种基于改进神经过程模型的缺失数据填充算法，该算法可有效提升小数据集背景下的缺失数据填充性能。首先，将观测到的时间序列进行单一表示，由神经网络得到各自的表征向量；其次，通过神经过程模型获得数据的分布函数，并在训练阶段引入修正系数α，从而根据数据缺失率更加精确地确定训练数据的采样率；最后，加入填充过程，通过训练好的模型估计数据缺失值。为检验算法性能，在海洋表面温度数据集以及北京PM2.5含量数据集上进行仿真实验，结果表明该算法在小数据集背景下具有良好的填充效果。与其他算法相比，所提算法在高缺失率的情况下具有更低的均方根误差。
Missing data imputing is a serious problem in the field of data analysis and process, which is extremely intractable in the case of the small dataset especially. In view of this problem, a missing data imputing algorithm based on modified neural process is proposed, which can improve the imputing performance in the background of the small dataset. Firstly, the observed time series is single-represented and then obtain the symptomatic vector respectively through the neural network. Secondly, it can acquire the distribution function of the data via the neural process and introduce the correction coefficient α to determine the sampling rate more exactly based on missing rate in the training stage. Finally, it imported the imputing process and estimated the missing data via trained model. Experiments are carried out on the sea surface temperature dataset and the Beijing PM2.5 dataset to verify the performance of the algorithm. The experiments show that the algorithm has an excellent performance in the context of small datasets, and it has a lower root mean square error compared with other algorithms.
[1] Kreindler D M, Lumsden C J. The effects of the irregular sample and missing data in time series analysis[J]. Nonlinear Dynamics Systems Analysis for the Behavioral Sciences Using Real Data. Boca Raton:CRC Press, 2016:149-172.Psychology & Life Sciences, 2006, 10(2):187-214.
[2] Balouji E, Salor Ö, Ermis M. Exponential smoothing of multiple reference frame components with GPUs for real-time detection of time-varying harmonics and interharmonics of EAF currents[J]. IEEE Transactions on Industry Applications, 2018, 54(6):6566-6575.
[3] Kozera R, Wilkołazka M. Natural spline interpolation and exponential parameterization for length estimation of curves[C]//AIP Conference Proceedings. Rhodes:AIP Publishing, 2017, 1863(1):400010.
[4] Newsham G R, Birt B J. Building-level occupancy data to improve ARIMA-based electricity use forecasts[C]//Proceedings of the 2nd ACM Workshop on Embedded Sensing Systems for Energy-Efficiency in Building. Zurich:ACM, 2010:13-18.
[5] Lippi M, Bertini M, Frasconi P. Short-term traffic flow forecasting:an experimental comparison of time-series analysis and supervised learning[J]. IEEE Transactions on Intelligent Transportation Systems, 2013, 14(2):871-882.
[6] Wang J, De Vries A P, Reinders M J T. Unifying user-based and item-based collaborative filtering approaches by similarity fusion[C]//Proceedings of the 29th Annual International ACM SIGIR Conference on Research and Development in Information Retrieval. Boston:ACM, 2006:501-508.
[7] Yu H F, Rao N, Dhillon I. Temporal regularized matrix factorization for high-dimensional time series prediction[C]//Advances in Neural Information Processing Systems. Barcelona:NIPS, 2016:847-855.
[8] Hron K, Templ M, Filzmoser P. Imputation of missing values for compositional data using classical and robust methods[J]. Computational Statistics & Data Analysis, 2010, 54(12):3095-3107.
[9] Stekhoven D J, Bühlmann P. MissForest:non-parametric missing value imputation for mixed-type data[J]. Bioinformatics, 2012, 28(1):112-118.
[10] Jia Z J, Song T W, Wang J X, et al. A time-series missing data completion method based on Fourier transform and kNNI algorithm[J]. Software Engineering, 2017, 20(3):9-13.
[11] Miller D, Ward A, Bambos N, et al. Physiological waveform imputation of missing data using convolutional autoencoders[C]//2018 IEEE 20th International Conference on e-Health Networking, Applications and Services (Healthcom). Ostrawa:IEEE, 2018:1-6.
[12] Wang H, Yuan Z L, Chen Y B, et al. An industrial missing values processing method based on generating model[J]. Computer Networks, 2019, 158:61-68.
[13] Duan Y J, Lü Y S, Kang W W, et al. A deep learning based approach for traffic data imputation[C]//17th International IEEE Conference on Intelligent Transportation Systems (ITSC). Qingdao:IEEE, 2014:912-917.
[14] Garnelo M, Schwarz J, Rosenbaum D, et al. Neural processes[J]. arXiv preprint arXiv:1807.01622, 2018.
[15] Wentz F J, Gentemann C, Smith D, et al. Satellite measurements of sea surface temperature through clouds[J]. Science, 2000, 288(5467):847-850.
[16] NOAA/Pacific Marine Environmental Laboratory. Tropical atmosphere ocean[DB/OL].[2019-06-06]. http://www.pmel.noaa.gov/tao/proj_over/proj_over.html.
[17] Dua D, Graff C. UCI Machine learning repository[DB/OL].[2019-07-30]. https://archive.ics.uci.edu/ml/datasets/Beijing+PM2.5+Data.
[18] Zhang Z L, Rao B D. Sparse signal recovery with temporally correlated source vectors using sparse Bayesian learning[J]. IEEE Journal of Selected Topics in Signal Processing, 2011, 5(5):912-926.
[19] Strauman A S, Bianchi F M, Mikalsen K, et al. Classification of postoperative surgical site infections from blood measurements with missing data using recurrent neural networks[C]//2018 IEEE EMBS International Conference on Biomedical & Health Informatics (BHI). Las Vegas:IEEE, 2018:307-310.
