I takt med internets utveckling blir det allt lättare för människor att ta informerade beslut i deras vardag. Detta gäller allt ifrån husköp till inköp av den mest prisvärda matkassen. Företag som agerar inom en abonnemangsbaserad marknad måste arbeta desto hårdare för att behålla sina kunder då det är lättare än någonsin att sprida sina upplevelser från olika företag, både bra och dåliga. Den här uppsatsen analyserar hur utskick ett företags CRM-avdelning gör påverkar huruvida kunderna väljer att avsluta sin tjänst (CRM står för Customer Relationship Management). Det visar sig att olika typer av säljkommunikation bidrar till att behålla kunder, medan majoriteten av informationskommunikationen bidrar till att öka andelen som avslutar sitt abonnemang. Det argumenteras under uppsatsens gång att en mer detaljerad taggning av de kunder som tagit del av olika kommunikationer kan bidra till en förbättrad analys, och i slutändan en förbättrad process i förebyggandet av avslut.

Den här rapporten är en pilotstudie på uppdrag av Trafikverket, där tågtider undersöks. Trafikverket mäter ankomst- och avgångstider genom en maskinellt uppmätt punktlighetsskattning på när tågen ankommer till, samt avgår från, perrongen. Förstudiens syfte var att undersöka om punktlighetsskattningen överensstämmer med de faktiska ankomst- och avgångstiderna. Det undersöks även vad som påverkar tågs eventuella förseningar genom att studera tidsskillnaderna mellan de planerande tiderna och de faktiska tiderna. Urvalet bestod av 251 observationer för platserna Uppsala Centralstation och Södertälje syd, som undersöktes på sekundnivå. Ytterligare utvidgades studien med att lägga till Gävle Centralstation som undersöker trunkerade tider på minutnivå, urvalet blev då totalt 282 observationer. För att besvara studiens frågeställningar användes multipel linjär regression samt ett illustrativt regressionsträd. Variablerna som ingick i studien förklarade inte modellen i hög utsträckning, men resultaten visade att det var en signifikant skillnad mellan de olika uppmätta tiderna för tågen att ankomma till, samt avgå från, perrongen jämfört med dess faktiska tider.

Previous studies have found that the socioeconomic risk factors education, income and family type were independently associated with mortality after first myocardial infarction (MI). In this population-based cohort study we study how these socioeconomic risk factors are related to the mortality among first MI survivors within 365 days. To this end, we use the four-way decomposition analysis which decomposes the total effect of the exposure into the effect that is due to only mediation, due to only interaction, due to both mediation and interaction and due to neither mediation nor interaction. Using education as exposure and income or family type as mediator two different models are estimated (adjusted for gender, age, region of birth and year of admission). For the model in which income was used as mediator, results show that individuals with lower education and a lower income are at greater risk of mortality, compared to individuals with the highest income level. The mediated effect of income on mortality decreases as the income approaches the highest income. In the model with family type as mediator the mediated effect on mortality was found to be low, meanwhile the direct effect of lower education was most substantial to the total risk of mortality.

In this thesis some aspects of process capability analysis are considered. Process capability analysis deals with how to assess the capability of manufacturing processes. Based on the process capability analysis one can determine how the process performs relative to its product requirements or specifications. An important part within process capability analysis is the use of process capability indices. This thesis focuses on process capability indices in the situation when the specification limits are one- sided. The thesis consists of a summary and three papers, of which one is already published in an international journal. The summary gives a background to the research area, a short overview of the three papers, and some suggestions for future research. In Paper I, the frequency and use of process capability analysis together with statistical process control and design of experiments, within Swedish companies hiring alumni students are investigated. We also investigate what motivates organisations to implement or not implement these statistical methods, if there are differences in use that can be related to organisational types and what will be needed to increase the use. One conclusion drawn from the results is that the students employed in the Swedish industrial sector witness a modest use of these statistical methods and use in other sectors hiring the alumni is uncommon. In Paper II we present a graphical method useful when doing capability analysis having one-sided specification limits. This is an extension of the process capability plots previously developed for two-sided specification intervals. Under the assumption of normality we suggest estimated process capability plots to be used to assess process capability at a given significance level. The presented graphical approach is helpful to determine if it is the variability, the deviation from target, or both that need to be reduced to improve the capability. In Paper III the situation with non-negative process data having a skew distribution with a long tail towards large values are considered, when an upper specification limit only exists and the target value is 0. No proper indices exist for this specific situation, which is common in practice. We contribute to this area by proposing a new class of indices designed for skew, zero-bound distributions when an upper specification only exists and the target value is 0. This new class of indices is simple and possesses properties desirable for process capability indices. Two estimators of the proposed index are studied and the asymptotic distributions of these estimators are derived. Furthermore, we consider decision procedures, based on the estimated indices, suitable for deeming the process capability or not. / <p>Godkänd; 2006; 20061201 (haneit)</p>

No description available.

This thesis consists of 4 papers, presented in Paper A-D, on particle- based online smoothing and parameter inference in general state-space hidden Markov models. In Paper A a novel algorithm, the particle-based, rapid incremental smoother (PaRIS), aimed at efficiently performing online approxima- tion of smoothed expectations of additive state functionals in general hidden Markov models, is presented. The algorithm has, under weak assumptions, linear computational complexity and very limited mem- ory requirements. The algorithm is also furnished with a number of convergence results, including a central limit theorem. In Paper B the problem of marginal smoothing in general hidden Markov models is tackled. A novel, PaRIS-based algorithm is presented where the marginal smoothing distributions are approximated using a lagged estimator where the lag is set adaptively. In Paper C an estimator of the tangent filter is constructed, yield- ing in turn an estimator of the score function. The resulting algorithm is furnished with theoretical results, including a central limit theorem with a uniformly bounded variance. The resulting estimator is applied to online parameter estimation via recursive maximum liklihood. Paper D focuses on the problem of online estimation of parameters in general hidden Markov models. The algorithm is based on a for- ward implementation of the classical expectation-maximization algo- rithm. The algorithm uses the PaRIS algorithm to achieve an efficient algorithm. / Denna avhandling består av fyra artiklar, presenterade i Paper A-D, som behandlar partikelbaserad online-glättning och parameter- skattning i generella dolda Markovkedjor. I papper A presenteras en ny algoritm, PaRIS, med målet att effek- tivt beräkna partikelbaserade online-skattningar av glättade väntevär- den av additiva tillståndsfunktionaler. Algoritmen har, under svaga villkor, en beräkningskomplexitet som växer endast linjärt med antalet partiklar samt högst begränsade minneskrav. Dessutom härleds ett an- tal konvergensresultat för denna algoritm, såsom en central gränsvärdes- sats. Algoritmen testas i en simuleringsstudie. I papper B studeras problemet att skatta marginalglättningsfördel- ningen i dolda Markovkedjor. Detta åstadkoms genom att exekvera PaRIS-algoritmen i marginalläge. Genom ett argument om mixning i Markovkedjor motiveras att avbryta uppdateringen efter en av ett stoppkriterium bestämd fördröjning vilket ger en adaptiv fördröjnings- glättare. I papper C studeras problemet att beräkna derivator av filterfördel- ningen. Dessa används för att beräkna gradienten av log-likelihood funktionen. Algoritmen, som innehåller en uppdateringsmekanism lik- nande den i PaRIS, förses med ett antal konvergensresultat, såsom en central gränsvärdessats med en varians som är likformigt begränsad. Den resulterande algoritmen används för att konstruera en rekursiv parameterskattningsalgoritm. Papper D fokuserar på online-estimering av modellparametrar i generella dolda Markovkedjor. Den presenterade algoritmen kan ses som en kombination av PaRIS algoritmen och en nyligen föreslagen online-implementation av den klassiska EM-algoritmen. / <p>QC 20171009</p>

No description available.

The development of the 5G network is in progress, and one part of the process that needs to be optimised is the handover. This operation, consisting of changing the base station (BS) providing data to a user equipment (UE), needs to be efficient enough to be a seamless operation. From the BS point of view, this operation should be as economical as possible, while satisfying the UE needs.  In this thesis, the problem of 5G handover has been addressed, and the chosen tool to solve this problem is reinforcement learning. A review of the different methods proposed by reinforcement learning led to the restricted field of model-free, off-policy methods, more specifically the Q-Learning algorithm. On its basic form, and used with simulated data, this method allows to get information on which kind of reward and which kinds of action-space and state-space produce good results. However, despite working on some restricted datasets, this algorithm does not scale well due to lengthy computation times. It means that the agent trained can not use a lot of data for its learning process, and both state-space and action-space can not be extended a lot, restricting the use of the basic Q-Learning algorithm to discrete variables. Since the strength of the signal (RSRP), which is of high interest to match the UE needs, is a continuous variable, a continuous form of the Q-learning needs to be used. A function approximation method is then investigated, namely artificial neural networks. In addition to the lengthy computational time, the results obtained are not convincing yet. Thus, despite some interesting results obtained from the basic form of the Q-Learning algorithm, the extension to the continuous case has not been successful. Moreover, the computation times make the use of reinforcement learning applicable in our domain only for really powerful computers.

No description available.

No description available.