New methods for time-to-event prediction are proposed by extending the Cox proportional hazards model with neural networks. Building on methodology from nested case-control studies, we propose a loss function that scales well to large data sets, and enables fitting of both non-proportional extensions model. Through simulation is verified be good approximation partial log-likelihood. The compared existing methodologies real-world found highly competitive, typically yielding best performance...

Many banks and credit institutions are required to assess the value of dwellings in their mortgage portfolio. This valuation often relies on an Automated Valuation Model (AVM). Moreover, these report models accuracy by two numbers: The fraction predictions within ±20% ±10% range from true values. Until recently, AVMs tended be hedonic regression models, but lately machine learning approaches like random forest gradient boosted trees have been increasingly applied. Both traditional rely...

We introduce Locally Interpretable Tree Boosting (LitBoost), a tree boosting model tailored to applications where the data comes from several heterogeneous yet known groups with limited number of observations per group. LitBoost constraints complexity Gradient Boosted Trees in way that allows us express final as set local Generalized Additive Models, yielding significant interpretability benefits while still maintaining some predictive power model. use house price prediction motivating...

Missing values are problematic for the analysis of microarray data. Imputation methods have been compared in terms similarity between imputed and true simulation experiments not their influence on final analysis. The focus has missing at random, while entries also random.We investigate imputation detection differentially expressed genes from cDNA We apply ANOVA microarrays SAM look to that lost because imputation. show this new measure provides useful information traditional root mean...

Summary Climate change will affect the insurance industry. We develop a Bayesian hierarchical statistical approach to explain and predict losses due weather events at local geographic scale. The number of weather-related claims is modelled by combining generalized linear models with spatially smoothed variable selection. Using Gibbs sampling reversible jump Markov chain Monte Carlo methods, this model fitted on daily data from each 319 municipalities which constitute southern central Norway...

Infectious salmon anemia (ISA) is one of the main infectious diseases in Atlantic farming with major economical implications. Despite strong regulatory interventions, ISA epidemic not under control, worldwide. We study data covering Norway from 2002 to 2005 and propose a stochastic space-time model for transmission virus. seaway between farm sites, through shared management infrastructure, biomass effects other potential pathways within industry. find that has an effect on infectiousness,...

Abstract Objective To investigate item non-response in a postal food-frequency questionnaire (FFQ), and to assess the effect of substituting/imputing missing values on dietary intake levels Norwegian Women Cancer study (NOWAC). We have adapted probably for first time applied k nearest neighbours (KNN) imputation FFQ data. Design Data from recent reproducibility were used. The was mailed twice (test–retest) about 3 months apart same subjects. Missing responses test imputed using null value...

In order to assess the potential of regional climate models be used project future weather events, a first step is study model forced by actual weather, or more precisely reanalysis data. this paper we investigate how well Norwegian HIRHAM, ERA-40 data, compares observed precipitation data from Meteorological Institute over mainland. This aims show standard methods statistical testing may dynamic downscaling. Methods considered are Kolmogorov—Smirnov two-sample test, Fisher exact test for...

Preference data occur when assessors express comparative opinions about a set of items, by rating, ranking, pair comparing, liking, or clicking. The purpose preference learning is to ( a) infer on the shared consensus group users, sometimes called rank aggregation, b) estimate for each user her individual ranking indicates only incomplete preferences; latter an important part recommender systems. We provide overview probabilistic approaches learning, including Mallows, Plackett–Luce, and...

Abstract. Real‐world phenomena are frequently modelled by Bayesian hierarchical models. The building‐blocks in such models the distribution of each variable conditional on parent and/or neighbour variables graph. specifications centre and spread these distributions may be well motivated, whereas tail often left to convenience. However, posterior a parameter depend strongly arbitrary specifications. This is not easily detected complex In this article, we propose graphical diagnostic, Local...

This thesis proposes a general method for drawing inference on the main drivers of adoption processes taking place social graph structures, and, subsequently, predicting future adoptions.Inference and prediction are performed by assuming that individuals may adopt when influenced either external factors, peer-to-peer influence mechanisms, or both.The is tested real-world data, its performances found to be satisfactory.We proceed investigate method's robustness missing information pertaining...

Clicking data, which exists in abundance and contains objective user preference information, is widely used to produce personalized recommendations web-based applications. Current popular recommendation algorithms, typically based on matrix factorizations, often focus achieving high accuracy. While good clickthrough rates, diversity of the recommended items overlooked. Moreover, most algorithms do not interpretable uncertainty quantifications recommendations. In this work, we propose...

Adoptions of a new innovation such as product, service or idea are typically driven both by peer‐to‐peer social interactions and external influence. Social graphs usually used to efficiently model the interactions, where adopters influence their peers also adopt innovation. However, may spread through individuals close adopters, known tattlers, who only share information regarding We extend an inhomogeneous Poisson process accounting for include optional tattling stage, we term extension...

Abstract. Dynamical downscaling of earth system models is intended to produce high-resolution climate information at regional local scales. Current models, while adequate for describing temperature distributions relatively small scales, struggle when it comes precipitation distributions. In order better match the distribution observed over Norway, we consider approaches statistical adjustment output from a model forced with ERA-40 reanalysis boundary conditions. As second step, try correct...

Abstract The posterior predictive p value ( ) was invented as a Bayesian counterpart to classical values. methodology can be applied discrepancy measures involving both data and parameters can, hence, targeted check for various modeling assumptions. interpretation however, difficult since the distribution of under assumptions varies substantially between cases. A calibration procedure has been suggested, treating test statistic in prior test. In this paper, we suggest that may instead based...

We propose the Pseudo-Mallows distribution over set of all permutations $n$ items, to approximate posterior with a Mallows likelihood. The model has been proven be useful for recommender systems where it can used learn personal preferences from highly incomplete data provided by users. Inference based on MCMC is however slow, preventing its use in real time applications. product univariate discrete Mallows-like distributions, constrained remain space permutations. quality approximation...

Clicking data, which exists in abundance and contains objective user preference information, is widely used to produce personalized recommendations web-based applications. Current popular recommendation algorithms, typically based on matrix factorizations, often have high accuracy achieve good clickthrough rates. However, diversity of the recommended items, can greatly enhance experiences, overlooked. Moreover, most algorithms do not interpretable uncertainty quantifications recommendations....