ISBN
Formato digital
979-13-87837-54-9
Fecha de publicación
06-10-2025
Licencia
D. R. © Copyright 2025. Alma Y. Alanis, Jorge Galvez, Omar Avalos, Eduardo Méndez-Palos, Jorge D. Rios, Adriana Peña Perez-Negron & Gabriel Martínez Soltero
Todos los contenidos de esta obra se comparten bajo la licencia Creative Commons Atri-bución/Reconocimiento-NoComercial-CompartirIgual 4.0 Internacional (CC BY-NC-SA 4.0). Esto implica que no está autorizado el uso comercial de la obra original ni de las eventuales obras derivadas, las cuales deberán distribuirse bajo la misma licencia que rige la obra original. No obstante, se permite a terceros compartir el contenido siempre y cuando se reconozca debidamente la autoría y la publicación original en esta editorial.
Hannia Macías Hernandez
Universidad de Guadalajara
0009-0005-2346-9519
Alma Yolanda Alanis García
Universidad de Guadalajara
0000-0001-9600-779X
Eduardo Rangel Heras
Universidad de Guadalajara
Arturo Valdivia G
Universidad de Guadalajara
0000-0001-8472-1523
Óscar Didier Sánchez Sánchez
Universidad Autónoma de Guadalajara
0000-0001-8215-6348
Acerca de
Time series analysis involves identifying patterns, trends, and seasonal variations in data collected at specific intervals, with applications in finance, economics, and weather forecasting. A critical challenge in time series analysis is handling missing data, which can arise from non-response, errors, or equipment malfunctions. Missing data can bias results and reduce accuracy, necessitating effective techniques such as deletion, imputation (e.g., mean, regression, or multiple imputation), and algorithms capable of handling missing values natively. The choice of method depends on the nature of the missing data—Missing Completely at Random (MCAR), Missing at Random (MAR), or Missing Not at Random (MNAR). Advanced techniques like regressive, autoregressive, and vector autoregressive models have been employed for imputation in various domains, including healthcare and environmental monitoring.
This paper introduces a novel Deep Neural Network (DNN) approach for cleaning and filtering time series data, outperforming nine classical techniques such as linear interpolation, spline, and moving mean/median. The proposed DNN, a 1D convolutional autoencoder, processes noisy data directly without requiring preprocessing steps like outlier removal. Evaluated on temperature time series data from a compressor, the DNN achieved a Root Mean Square Error (RMSE) of 6°C, significantly lower than the best classical method (45°C). The DNN’s ability to accurately impute missing data and follow trends, even with large gaps, demonstrates its robustness and superiority over traditional methods. This study highlights the potential of DNNs in enhancing time series data analysis and imputation.
Referencias
Dan, E.L.; Dinsoreanu, M.; Muresan, R.C. Accuracy of Six Interpola- tion Methods Applied on Pupil Diameter Data. In Proceedings of the 2020 22nd IEEE International Conference on Automation, Quality and Testing, Robotics – THETA, AQTR 2020 – Proceedings; Institute of Electrical and Electronics Engineers Inc., May 1 2020.
Noor, N.M.; Al Bakri Abdullah, M.M.; Yahaya, A.S.; Ramli, N.A. Comparison of Linear Interpolation Method and Mean Method to Re- place the Missing Values in Environmental Data Set. In Proceedings of the Materials Science Forum; Trans Tech Publications Ltd, 2015; Vol. 803, pp. 278–281.
Zhang, Z. Missing Data Imputation: Focusing on Single Imputation.
Ann Transl Med 2016, 4, doi:10.3978/j.issn.2305-5839.2015.12.38.
Junger, W.L.; Ponce de Leon, A. Imputation of Missing Data in Time Series for Air Pollutants. Atmos Environ 2015, 102, 96–104, doi:10.1016/j.atmosenv.2014.11.049.
Murad, H.; Dankner, R.; Berlin, A.; Olmer, L.; Freedman, L.S. Imput- ing Missing Time-Dependent Covariate Values for the Discrete Time Cox Model. Stat Methods Med Res 2020, 29, 2074–2086, doi:10.1177/0962280219881168.
Liu, S.; Molenaar, P.C.M. IVAR: A Program for Imputing Missing Data in Multivariate Time Series Using Vector Autoregressive Mod- els. Behav Res Methods 2014, 46, 1138–1148, doi:10.3758/s13428-
014-0444-4.
Bashir, F.; Wei, H.L. Handling Missing Data in Multivariate Time Se- ries Using a Vector Autoregressive Model-Imputation (VAR-IM) Al- gorithm. Neurocomputing 2018, 276, 23–30, doi:10.1016/j.neucom.2017.03.097.
Dunsmuir, W.; Robinson, P.M. Estimation of Time Series Models in the Presence of Missing Data. J Am Stat Assoc 1981, 76, 560–568, doi:10.1080/01621459.1981.10477687.
Anava, O.; Hazan, E.; Zeevi, A. Online Time Series Prediction with Missing Data. In Proceedings of the International Confer- ence on MachineLearning, Lille, France; Lille, France, 2015; pp. 1– 9.
Pedreschi, R.; Hertog, M.L.A.T.M.; Carpentier, S.C.; Lammertyn, J.; Robben, J.; Noben, J.P.; Panis, B.; Swennen, R.; Nicolai, B.M. Treatment of Missing Values for Multivariate Statistical Analysis of Gel-Based Proteomics Data. Proteomics 2008, 8, 1371–1383, doi:10.1002/pmic.200700975.
Choong, M.K.; Charbit, M.; Yan, H. Autoregressive-Model-Based Missing Value Estimation for DNA Microarray Time Series Data. IEEE Transactions on Information Technology in Biomedicine 2009, 13, 131–137, doi:10.1109/TITB.2008.2007421.