liu.seSearch for publications in DiVA
Change search
Link to record
Permanent link

Direct link
Publications (3 of 3) Show all publications
Svahn, C. (2022). Prediction Methods for High Dimensional Data with Censored Covariates. (Doctoral dissertation). Linköping: Linköping University Electronic Press
Open this publication in new window or tab >>Prediction Methods for High Dimensional Data with Censored Covariates
2022 (English)Doctoral thesis, comprehensive summary (Other academic)
Alternative title[sv]
Prediktionsmetoder för högdimensionella data med censurerade kovariater
Abstract [en]

While access to data steadily increases, not all data are straight-forward to use for prediction. Censored data are common in several industrial scenarios, and typically arise when there are some limitations to measuring equipment such as for instance concentration measuring equipment in chemistry or signal receivers in signal processing. 

In this thesis, we take several angles to censored covariate data for prediction problem. We explore the impact on both covariates and the response when the censored covariates are imputed. We consider linear approaches as well as non-linear approaches, and we explore how both frequentist models as well as Bayesian models perform with censored covariate data. While the focus is using the imputed covariate data for prediction, we also investigate model parameter inference and uncertainty inferred by the imputations. 

We use real, censored covariate telecommunications data for prediction with some of the most commonly used prediction models and evaluate the performance when single imputations are made. We propose a selective multiple imputation approach which is suitable for high dimensional data that perform well with heavy censoring. We take a Bayesian linear regression approach leveraging information from auxiliary variables using multivariate regression and introduce multivariate draws from conditional distributions to update censored values in the covariates. We fnally offer a bridge between the fexibility of Neural Networks and the probabilistic nature of Bayesian methods by taking a Variational Autoencoder approach and introducing Zero-Infated Truncated Gaussian likelihoods for the covariates to better ft the censored distributions. 

Abstract [sv]

I många industriella sammanhang finns stora mängder data att tillgå. Dessa data är dock ofta inkompletta, och strategier behövs för kunna nyttja data på bästa sätt när de används för prediktion. Mycket forskning har fortgått för att hantera saknade data i responsvariabeln, den variabel som ska predikteras, medan mindre forskning inriktats på saknade värden i kovariater, variablerna som används för att prediktera responsvariabeln. Ännu mindre forskning har fokuserat på så kallade censurerade data. Censurerade data är ett specialfall av saknade data där data är partiellt observerat, men som inte kan observeras fullt då exempelvis värden under en specifik tröskel inte går att mäta. Detta är vanligt i exempelvis signaldata, där mottagaren av signalen har en undre gräns för hörbarhet.

I denna avhandling bidrar vi till forskning för censurerade kovariater i prediktionsmodeller genom att introducera strategier som är snabbare och kan hantera mer komplexa beroenden i data än befintliga metoder. Vi angriper problemet från flertalet vinklar, och detta arbete presenterar metoder för att både kunna prediktera data, återställa de censurerade värdena och parametrar från datagenereringsprocessen med god precision.

Vi ställer olika traditionella metoder mot varandra och utvärderar hur enkla metoder för att ersätta, så kallat imputera, censurerade värden påverkar osäkerheten i prediktioner och presenterar alternativ till att ta specifika beslut under stor osäkerhet. Vi visar att det kan vara en fördel att vid tung censurering inte imputera alla censurerade värden och på så sätt åstadkomma kortare beräkningstider. Vi presenterar hur man kan använda beroenden mellan kovariater för att åstadkomma mer effektiva beräkningar och mer precisa imputationer. Slutligen visar vi hur man kan ändra antaganden för sannolikhetsfördelningarna för censorerad data för att kunna imputera med bättre precision. Vi gör detta med en metod som är snabb, flexibel för komplexa data och som kan generera skattningar på osäkerhet.

Place, publisher, year, edition, pages
Linköping: Linköping University Electronic Press, 2022. p. 26
Series
Linköping Studies in Arts and Sciences, ISSN 0282-9800 ; 839Linköping Studies in Statistics, ISSN 1651-1700 ; 16
Keywords
statistics, machine learning, censored covariates, statistik, maskininlärning, censurerade kovariater
National Category
Probability Theory and Statistics
Identifiers
urn:nbn:se:liu:diva-187763 (URN)10.3384/9789179293994 (DOI)9789179293987 (ISBN)9789179293994 (ISBN)
Public defence
2022-10-04, Ada Lovelace, B-huset, Campus Valla, Linköping, 13:15
Opponent
Supervisors
Available from: 2022-08-23 Created: 2022-08-23 Last updated: 2022-09-08Bibliographically approved
Nielsen, K., Svahn, C., Rodriguez Déniz, H. & Hendeby, G. (2021). UKF Parameter Tuning for Local Variation Smoothing. In: Proceedings of the 2021 IEEE International Conference on Multisensor Fusion and Integration for Intelligent Systems (MFI): . Paper presented at 2021 IEEE International Conference on Multisensor Fusion and Integration for Intelligent Systems (MFI), Karlsruhe, Germany, 23-25 September 2021. Institute of Electrical and Electronics Engineers (IEEE)
Open this publication in new window or tab >>UKF Parameter Tuning for Local Variation Smoothing
2021 (English)In: Proceedings of the 2021 IEEE International Conference on Multisensor Fusion and Integration for Intelligent Systems (MFI), Institute of Electrical and Electronics Engineers (IEEE), 2021Conference paper, Published paper (Refereed)
Abstract [en]

The unscented Kalman filter (UKF) is a method to solve nonlinear dynamic filtering problems, which internally uses the unscented transform (UT). The behavior of the UT is controlled by design parameters, seldom changed from the values suggested in early UT/UKF publications. Despite the knowledge that the UKF can perform poorly when the parameters are improperly chosen, there exist no wide spread intuitive guidelines for how to tune them. With an application relevant example, this paper shows that standard parameter values can be far from optimal. By analyzing how each parameter affects the resulting UT estimate, guidelines for how the parameter values should be chosen are developed. The guidelines are verified both in simulations and on real data collected in an underground mine. A strategy to automatically tune the parameters in a state estimation setting is presented, resulting in parameter values inline with developed guidelines.

Place, publisher, year, edition, pages
Institute of Electrical and Electronics Engineers (IEEE), 2021
Keywords
unscented Kalman filter, auto-tuning, WASP_publications
National Category
Control Engineering
Identifiers
urn:nbn:se:liu:diva-183295 (URN)10.1109/MFI52462.2021.9591188 (DOI)000853882500029 ()9781665445214 (ISBN)9781665445221 (ISBN)
Conference
2021 IEEE International Conference on Multisensor Fusion and Integration for Intelligent Systems (MFI), Karlsruhe, Germany, 23-25 September 2021
Funder
Wallenberg AI, Autonomous Systems and Software Program (WASP)
Note

Funding: Wallenberg AI, Autonomous Systems and Software Program (WASP) - Knut and Alice Wallenberg Foundation

Available from: 2022-03-01 Created: 2022-03-01 Last updated: 2023-04-20
Svahn, C., Sysoev, O., Cirkic, M., Gustafsson, F. & Berglund, J. (2019). Inter-Frequency Radio Signal Quality Prediction for Handover, Evaluated in 3GPP LTE. In: 2019 IEEE 89th Vehicular Technology Conference (VTC2019-Spring): . Paper presented at 2019 IEEE 89th Vehicular Technology Conference (VTC2019-Spring), Kuala Lumpur, Malaysia, Malaysia, 28 April-1 May 2019 (pp. 1-5). IEEE
Open this publication in new window or tab >>Inter-Frequency Radio Signal Quality Prediction for Handover, Evaluated in 3GPP LTE
Show others...
2019 (English)In: 2019 IEEE 89th Vehicular Technology Conference (VTC2019-Spring), IEEE, 2019, p. 1-5Conference paper, Published paper (Refereed)
Abstract [en]

Radio resource management in cellular networks is typically based on device measurements reported to the serving base station. Frequent measuring of signal quality on available frequencies would allow for highly reliable networks and optimal connection at all times. However, these measurements are associated with costs, such as dedicated device time for performing measurements when the device will be unavailable for communication. To reduce the costs, we consider predictions of inter-frequency radio quality measurements that are useful to assess potential inter-frequency handover decisions. In this contribution, we have considered measurements from a live 3GPP LTE network. We demonstrate that straightforward applications of the most commonly used machine learning models are unable to provide high accuracy predictions. Instead, we propose a novel approach with a duo-threshold for high accuracy decision recommendations. Our approach leads to class specific prediction accuracies as high as 92% and 95%, still drastically reducing the need for inter-frequency measurements.

Place, publisher, year, edition, pages
IEEE, 2019
Series
IEEE VEHICULAR TECHNOLOGY CONFERENCE, ISSN 1550-2252
National Category
Computer Sciences Communication Systems
Identifiers
urn:nbn:se:liu:diva-159307 (URN)10.1109/VTCSpring.2019.8746369 (DOI)000482655600080 ()2-s2.0-85068961000 (Scopus ID)978-1-7281-1217-6 (ISBN)978-1-7281-1218-3 (ISBN)
Conference
2019 IEEE 89th Vehicular Technology Conference (VTC2019-Spring), Kuala Lumpur, Malaysia, Malaysia, 28 April-1 May 2019
Funder
Wallenberg AI, Autonomous Systems and Software Program (WASP)
Note

Funding agencies: Wallenberg AI, Autonomous Systems and Software Program (WASP) - Knut and Alice Wallenberg Foundation

Available from: 2019-08-06 Created: 2019-08-06 Last updated: 2025-11-17Bibliographically approved
Organisations
Identifiers
ORCID iD: ORCID iD iconorcid.org/0000-0003-4271-6683

Search in DiVA

Show all publications