ResearchPad - Applied Mathematics https://www.researchpad.co Default RSS Feed en-us © 2020 Newgen KnowledgeWorks <![CDATA[LoAdaBoost: Loss-based AdaBoost federated machine learning with reduced computational complexity on IID and non-IID intensive care data]]> https://www.researchpad.co/product?articleinfo=Na533cb35-b26a-447b-bd62-8e125a165db4

Intensive care data are valuable for improvement of health care, policy making and many other purposes. Vast amount of such data are stored in different locations, on many different devices and in different data silos. Sharing data among different sources is a big challenge due to regulatory, operational and security reasons. One potential solution is federated machine learning, which is a method that sends machine learning algorithms simultaneously to all data sources, trains models in each source and aggregates the learned models. This strategy allows utilization of valuable data without moving them. One challenge in applying federated machine learning is the possibly different distributions of data from diverse sources. To tackle this problem, we proposed an adaptive boosting method named LoAdaBoost that increases the efficiency of federated machine learning. Using intensive care unit data from hospitals, we investigated the performance of learning in IID and non-IID data distribution scenarios, and showed that the proposed LoAdaBoost method achieved higher predictive accuracy with lower computational complexity than the baseline method.

]]>
<![CDATA[Exact flow of particles using for state estimations in unmanned aerial systems` navigation]]> https://www.researchpad.co/product?articleinfo=Nb8d1b185-24ca-4749-9cc9-bbc7ade34d0a

The navigation is a substantial issue in the field of robotics. Simultaneous Localization and Mapping (SLAM) is a principle for many autonomous navigation applications, particularly in the Global Navigation Satellite System (GNSS) denied environments. Many SLAM methods made substantial contributions to improve its accuracy, cost, and efficiency. Still, it is a considerable challenge to manage robust SLAM, and there exist several attempts to find better estimation algorithms for it. In this research, we proposed a novel Bayesian filtering based Airborne SLAM structure for the first time in the literature. We also presented the mathematical background of the algorithm, and the SLAM model of an autonomous aerial vehicle. Simulation results emphasize that the new Airborne SLAM performance with the exact flow of particles using for recursive state estimations superior to other approaches emerged before, in terms of accuracy and speed of convergence. Nevertheless, its computational complexity may cause real-time application concerns, particularly in high-dimensional state spaces. However, in Airborne SLAM, it can be preferred in the measurement environments that use low uncertainty sensors because it gives more successful results by eliminating the problem of degeneration seen in the particle filter structure.

]]>
<![CDATA[Predicting 30-day hospital readmissions using artificial neural networks with medical code embedding]]> https://www.researchpad.co/product?articleinfo=N1f40719a-4631-45e6-bedb-5cf8a42ecf53

Reducing unplanned readmissions is a major focus of current hospital quality efforts. In order to avoid unfair penalization, administrators and policymakers use prediction models to adjust for the performance of hospitals from healthcare claims data. Regression-based models are a commonly utilized method for such risk-standardization across hospitals; however, these models often suffer in accuracy. In this study we, compare four prediction models for unplanned patient readmission for patients hospitalized with acute myocardial infarction (AMI), congestive health failure (HF), and pneumonia (PNA) within the Nationwide Readmissions Database in 2014. We evaluated hierarchical logistic regression and compared its performance with gradient boosting and two models that utilize artificial neural networks. We show that unsupervised Global Vector for Word Representations embedding representations of administrative claims data combined with artificial neural network classification models improves prediction of 30-day readmission. Our best models increased the AUC for prediction of 30-day readmissions from 0.68 to 0.72 for AMI, 0.60 to 0.64 for HF, and 0.63 to 0.68 for PNA compared to hierarchical logistic regression. Furthermore, risk-standardized hospital readmission rates calculated from our artificial neural network model that employed embeddings led to reclassification of approximately 10% of hospitals across categories of hospital performance. This finding suggests that prediction models that incorporate new methods classify hospitals differently than traditional regression-based approaches and that their role in assessing hospital performance warrants further investigation.

]]>
<![CDATA[Topological data analysis of zebrafish patterns]]> https://www.researchpad.co/product?articleinfo=Nd36d9c09-878e-4ada-b6cb-8e6031766d16

Significance

While pattern formation has been studied extensively using experiments and mathematical models, methods for quantifying self-organization are limited to manual inspection or global measures in many applications. Our work introduces a methodology for automatically quantifying patterns that arise due to agent interactions. We combine topological data analysis and machine learning to provide a collection of summary statistics describing patterns on both microscopic and macroscopic scales. We apply our methodology to study zebrafish patterns across thousands of model simulations, allowing us to make quantitative predictions about the types of pattern variability present in wild-type and mutant zebrafish. Our work helps address the widespread challenge of quantifying agent-based patterns and opens up possibilities for large-scale analysis of biological data and mathematical models.

]]>
<![CDATA[Is it time to stop sweeping data cleaning under the carpet? A novel algorithm for outlier management in growth data]]> https://www.researchpad.co/product?articleinfo=N6ac4201b-e1d9-4dac-b706-1c6b88e127a6

All data are prone to error and require data cleaning prior to analysis. An important example is longitudinal growth data, for which there are no universally agreed standard methods for identifying and removing implausible values and many existing methods have limitations that restrict their usage across different domains. A decision-making algorithm that modified or deleted growth measurements based on a combination of pre-defined cut-offs and logic rules was designed. Five data cleaning methods for growth were tested with and without the addition of the algorithm and applied to five different longitudinal growth datasets: four uncleaned canine weight or height datasets and one pre-cleaned human weight dataset with randomly simulated errors. Prior to the addition of the algorithm, data cleaning based on non-linear mixed effects models was the most effective in all datasets and had on average a minimum of 26.00% higher sensitivity and 0.12% higher specificity than other methods. Data cleaning methods using the algorithm had improved data preservation and were capable of correcting simulated errors according to the gold standard; returning a value to its original state prior to error simulation. The algorithm improved the performance of all data cleaning methods and increased the average sensitivity and specificity of the non-linear mixed effects model method by 7.68% and 0.42% respectively. Using non-linear mixed effects models combined with the algorithm to clean data allows individual growth trajectories to vary from the population by using repeated longitudinal measurements, identifies consecutive errors or those within the first data entry, avoids the requirement for a minimum number of data entries, preserves data where possible by correcting errors rather than deleting them and removes duplications intelligently. This algorithm is broadly applicable to data cleaning anthropometric data in different mammalian species and could be adapted for use in a range of other domains.

]]>
<![CDATA[Exponential random graph model parameter estimation for very large directed networks]]> https://www.researchpad.co/product?articleinfo=N437fb42a-ebf8-44aa-9399-d12b1354408e

Exponential random graph models (ERGMs) are widely used for modeling social networks observed at one point in time. However the computational difficulty of ERGM parameter estimation has limited the practical application of this class of models to relatively small networks, up to a few thousand nodes at most, with usually only a few hundred nodes or fewer. In the case of undirected networks, snowball sampling can be used to find ERGM parameter estimates of larger networks via network samples, and recently published improvements in ERGM network distribution sampling and ERGM estimation algorithms have allowed ERGM parameter estimates of undirected networks with over one hundred thousand nodes to be made. However the implementations of these algorithms to date have been limited in their scalability, and also restricted to undirected networks. Here we describe an implementation of the recently published Equilibrium Expectation (EE) algorithm for ERGM parameter estimation of large directed networks. We test it on some simulated networks, and demonstrate its application to an online social network with over 1.6 million nodes.

]]>
<![CDATA[pyKNEEr: An image analysis workflow for open and reproducible research on femoral knee cartilage]]> https://www.researchpad.co/product?articleinfo=N0686bd46-1746-4f66-8610-270f1b75b482

Transparent research in musculoskeletal imaging is fundamental to reliably investigate diseases such as knee osteoarthritis (OA), a chronic disease impairing femoral knee cartilage. To study cartilage degeneration, researchers have developed algorithms to segment femoral knee cartilage from magnetic resonance (MR) images and to measure cartilage morphology and relaxometry. The majority of these algorithms are not publicly available or require advanced programming skills to be compiled and run. However, to accelerate discoveries and findings, it is crucial to have open and reproducible workflows. We present pyKNEEr, a framework for open and reproducible research on femoral knee cartilage from MR images. pyKNEEr is written in python, uses Jupyter notebook as a user interface, and is available on GitHub with a GNU GPLv3 license. It is composed of three modules: 1) image preprocessing to standardize spatial and intensity characteristics; 2) femoral knee cartilage segmentation for intersubject, multimodal, and longitudinal acquisitions; and 3) analysis of cartilage morphology and relaxometry. Each module contains one or more Jupyter notebooks with narrative, code, visualizations, and dependencies to reproduce computational environments. pyKNEEr facilitates transparent image-based research of femoral knee cartilage because of its ease of installation and use, and its versatility for publication and sharing among researchers. Finally, due to its modular structure, pyKNEEr favors code extension and algorithm comparison. We tested our reproducible workflows with experiments that also constitute an example of transparent research with pyKNEEr, and we compared pyKNEEr performances to existing algorithms in literature review visualizations. We provide links to executed notebooks and executable environments for immediate reproducibility of our findings.

]]>
<![CDATA[Disease-relevant mutations alter amino acid co-evolution networks in the second nucleotide binding domain of CFTR]]> https://www.researchpad.co/product?articleinfo=N211c75a7-eaac-4644-b655-cac4e239c2e4

Cystic Fibrosis (CF) is an inherited disease caused by mutations in the cystic fibrosis transmembrane conductance regulator (CFTR) ion channel. Mutations in CFTR cause impaired chloride ion transport in the epithelial tissues of patients leading to cardiopulmonary decline and pancreatic insufficiency in the most severely affected patients. CFTR is composed of twelve membrane-spanning domains, two nucleotide-binding domains (NBDs), and a regulatory domain. The most common mutation in CFTR is a deletion of phenylalanine at position 508 (ΔF508) in NBD1. Previous research has primarily concentrated on the structure and dynamics of the NBD1 domain; However numerous pathological mutations have also been found in the lesser-studied NBD2 domain. We have investigated the amino acid co-evolved network of interactions in NBD2, and the changes that occur in that network upon the introduction of CF and CF-related mutations (S1251N(T), S1235R, D1270N, N1303K(T)). Extensive coupling between the α- and β-subdomains were identified with residues in, or near Walker A, Walker B, H-loop and C-loop motifs. Alterations in the predicted residue network varied from moderate for the S1251T perturbation to more severe for N1303T. The S1235R and D1270N networks varied greatly compared to the wildtype, but these CF mutations only affect ion transport preference and do not severely disrupt CFTR function, suggesting dynamic flexibility in the network of interactions in NBD2. Our results also suggest that inappropriate interactions between the β-subdomain and Q-loop could be detrimental. We also identified mutations predicted to stabilize the NBD2 residue network upon introduction of the CF and CF-related mutations, and these predicted mutations are scored as benign by the MUTPRED2 algorithm. Our results suggest the level of disruption of the co-evolution predictions of the amino acid networks in NBD2 does not have a straightforward correlation with the severity of the CF phenotypes observed.

]]>
<![CDATA[Height of overburden fracture based on key strata theory in longwall face]]> https://www.researchpad.co/product?articleinfo=Nb6c965ed-0040-4b7a-b381-dffd2122531d

Among the three overburden zones (the caving zone, the fracture zone, and the continuous deformation zone) in longwall coal mining, the continuous deformation zone is often considered to be continuous without cracks, so continuum mechanics can be used to calculate the subsidence of overburden strata. Longwall coal mining, however, will induce the generation of wide cracks in the surface and thus may cause the continuous deformation zone to fracture. In this paper, whether there are cracks in the continuous deformation zone as well as the height of overburden fracture in longwall face and the subsidence and deformation of strata of different fracture penetration ratios were studied by means of physical simulation, theoretical analysis and numerical simulation. The results show that: (1) Rock stratum starts to fracture as long as it has slightly subsided for only tens of millimeters, and the height of fracture development is the height of working face overburden. (2) With the increase of fracture penetration ratio, the subsidence of key strata remains basically unchanged; the surface deformation range and the maximum compression deformation decrease, while the maximum horizontal movement and maximum horizontal tensile deformation increase. Therefore, the subsidence of overburden strata which have fractured but have not broken can be calculated through the continuum mechanics method.

]]>
<![CDATA[Lean back and wait for the alarm? Testing an automated alarm system for nosocomial outbreaks to provide support for infection control professionals]]> https://www.researchpad.co/product?articleinfo=N4571fdc0-2a2e-4467-acc9-eeadc2652757

Introduction

Outbreaks of communicable diseases in hospitals need to be quickly detected in order to enable immediate control. The increasing digitalization of hospital data processing offers potential solutions for automated outbreak detection systems (AODS). Our goal was to assess a newly developed AODS.

Methods

Our AODS was based on the diagnostic results of routine clinical microbiological examinations. The system prospectively counted detections per bacterial pathogen over time for the years 2016 and 2017. The baseline data covers data from 2013–2015. The comparative analysis was based on six different mathematical algorithms (normal/Poisson and score prediction intervals, the early aberration reporting system, negative binomial CUSUMs, and the Farrington algorithm). The clusters automatically detected were then compared with the results of our manual outbreak detection system.

Results

During the analysis period, 14 different hospital outbreaks were detected as a result of conventional manual outbreak detection. Based on the pathogens’ overall incidence, outbreaks were divided into two categories: outbreaks with rarely detected pathogens (sporadic) and outbreaks with often detected pathogens (endemic). For outbreaks with sporadic pathogens, the detection rate of our AODS ranged from 83% to 100%. Every algorithm detected 6 of 7 outbreaks with a sporadic pathogen. The AODS identified outbreaks with an endemic pathogen were at a detection rate of 33% to 100%. For endemic pathogens, the results varied based on the epidemiological characteristics of each outbreak and pathogen.

Conclusion

AODS for hospitals based on routine microbiological data is feasible and can provide relevant benefits for infection control teams. It offers in-time automated notification of suspected pathogen clusters especially for sporadically occurring pathogens. However, outbreaks of endemically detected pathogens need further individual pathogen-specific and setting-specific adjustments.

]]>
<![CDATA[A graph-based algorithm for RNA-seq data normalization]]> https://www.researchpad.co/product?articleinfo=N0b813aa9-b155-4778-93ba-b0f37d26ae8a

The use of RNA-sequencing has garnered much attention in recent years for characterizing and understanding various biological systems. However, it remains a major challenge to gain insights from a large number of RNA-seq experiments collectively, due to the normalization problem. Normalization has been challenging due to an inherent circularity, requiring that RNA-seq data be normalized before any pattern of differential (or non-differential) expression can be ascertained; meanwhile, the prior knowledge of non-differential transcripts is crucial to the normalization process. Some methods have successfully overcome this problem by the assumption that most transcripts are not differentially expressed. However, when RNA-seq profiles become more abundant and heterogeneous, this assumption fails to hold, leading to erroneous normalization. We present a normalization procedure that does not rely on this assumption, nor prior knowledge about the reference transcripts. This algorithm is based on a graph constructed from intrinsic correlations among RNA-seq transcripts and seeks to identify a set of densely connected vertices as references. Application of this algorithm on our synthesized validation data showed that it could recover the reference transcripts with high precision, thus resulting in high-quality normalization. On a realistic data set from the ENCODE project, this algorithm gave good results and could finish in a reasonable time. These preliminary results imply that we may be able to break the long persisting circularity problem in RNA-seq normalization.

]]>
<![CDATA[New physiological bench test reproducing nocturnal breathing pattern of patients with sleep disordered breathing]]> https://www.researchpad.co/product?articleinfo=N13bd4ad3-60c6-4376-997d-f10f1c975c0e

Previous studies have shown that Automatic Positive Airway Pressure devices display different behaviors when connected to a bench using theoretical respiratory cycle scripts. However, these scripts are limited and do not simulate physiological behavior during the night. Our aim was to develop a physiological bench that is able to simulate patient breathing airflow by integrating polygraph data. We developed an algorithm analyzing polygraph data and transformed this information into digital inputs required by the bench hardware to reproduce a patient breathing profile on bench. The inputs are respectively the simulated respiratory muscular effort pressure input for an artificial lung and the sealed chamber pressure to regulate the Starling resistor. We did simulations on our bench for a total of 8 hours and 59 minutes for a breathing profile from the demonstration recording of a Nox T3 Sleep Monitor. The simulation performance results showed that in terms of relative peak-valley amplitude of each breathing cycle, simulated bench airflow was biased by only 1.48% ± 6.80% compared to estimated polygraph nasal airflow for a total of 6,479 breathing cycles. For total respiratory cycle time, the average bias ± one standard deviation was 0.000 ± 0.288 seconds. For patient apnea events, our bench simulation had a sensitivity of 84.7% and a positive predictive value equal to 90.3%, considering 149 apneas detected both in polygraph nasal simulated bench airflows. Our new physiological bench would allow personalizing APAP device selection to each patient by taking into account individual characteristics of a sleep breathing profile.

]]>
<![CDATA[Semblance: An empirical similarity kernel on probability spaces]]> https://www.researchpad.co/product?articleinfo=N5f89a685-fd97-4dcd-b1ce-0c4277feb5d6

Semblance uses empirical data distributions to improve the sensitivity and robustness in downstream machine learning algorithms.

]]>
<![CDATA[Markovian approaches to modeling intracellular reaction processes with molecular memory]]> https://www.researchpad.co/product?articleinfo=N5bc9cb86-31db-495a-95d5-6a5749800496

Significance

Modeling intracellular processes has long relied on the markovian assumption. However, as soon as a reactant interacts with its environment, molecular memory definitely exists and its effects cannot be neglected. Since the Markov theory cannot translate directly to modeling and analysis of nonmarkovian processes, this leads to many significant challenges. We develop a formulation, namely the stationary generalized chemical-master equation, to model intracellular processes with molecular memory. This formulation converts a nonmarkovian question to a markovian one while keeping the stationary probabilistic behavior unchanged. Both a stationary generalized Fokker–Planck equation and a generalized linear noise approximation are further developed for the fast evaluation of fluctuations. These formulations can have broad applications and may help us discover new biological knowledge.

]]>
<![CDATA[Variable weights theory and its application to multi-attribute group decision making with intuitionistic fuzzy numbers on determining decision maker’s weights]]> https://www.researchpad.co/product?articleinfo=5c89771bd5eed0c4847d2486

The determination of the weights of decision makers (DMs) is an important problem in multi-attribute group decision making. Many approaches have been presented to determine DMs’ weights. However, the computed weight vectors of DMs are usually assumed to be constant in existing studies, and this may cause irrationalities in the decision results. Therefore, this article proposes a novel method to determine DMs’ weights based on variable weights theory in which the evaluation information is described as intuitionistic fuzzy sets (IFSs). First, DMs provide their assessment with IFSs, and the intuitionistic fuzzy weighted averaging (IFWA) operator is applied to obtain weighted decision matrix based on the prior given DMs’ and attributes’ weights. Second, the DMs’ weights are obtained based on variable weights theory, and an alternative decision can be computed. Finally, the converted value of the achieved IFS of each alternative is calculated, and the best appropriate alternative is acquired. Two illustrative examples and the comparisons with exsiting approaches are also used to reflect the effectiveness of the proposed approach.

]]>
<![CDATA[Assessing the role of transmission chains in the spread of HIV-1 among men who have sex with men in Quebec, Canada]]> https://www.researchpad.co/product?articleinfo=5c89773dd5eed0c4847d27bf

Background

Phylogenetics has been used to investigate HIV transmission among men who have sex with men. This study compares several methodologies to elucidate the role of transmission chains in the dynamics of HIV spread in Quebec, Canada.

Methods

The Quebec Human Immunodeficiency Virus (HIV) genotyping program database now includes viral sequences from close to 4,000 HIV-positive individuals classified as Men who have Sex with Men (MSMs), collected between 1996 and early 2016. Assessment of chain expansion may depend on the partitioning scheme used, and so, we produce estimates from several methods: the conventional Bayesian and maximum likelihood-bootstrap methods, in combination with a variety of schemes for applying a maximum distance criterion, and two other algorithms, DM-PhyClus, a Bayesian algorithm that produces a measure of uncertainty for proposed partitions, and the Gap Procedure, a fast non-phylogenetic approach. Sequences obtained from individuals in the Primary HIV Infection (PHI) stage serve to identify incident cases. We focus on the period ranging from January 1st 2012 to February 1st 2016.

Results and conclusion

The analyses reveal considerable overlap between chain estimates obtained from conventional methods, thus leading to similar estimates of recent temporal expansion. The Gap Procedure and DM-PhyClus suggest however moderately different chains. Nevertheless, all estimates stress that longer older chains are responsible for a sizeable proportion of the sampled incident cases among MSMs. Curbing the HIV epidemic will require strategies aimed specifically at preventing such growth.

]]>
<![CDATA[Normalization enhances brain network features that predict individual intelligence in children with epilepsy]]> https://www.researchpad.co/product?articleinfo=5c8823aad5eed0c484638dcf

Background and purpose

Architecture of the cerebral network has been shown to associate with IQ in children with epilepsy. However, subject-level prediction on this basis, a crucial step toward harnessing network analyses for the benefit of children with epilepsy, has yet to be achieved. We compared two network normalization strategies in terms of their ability to optimize subject-level inferences on the relationship between brain network architecture and brain function.

Materials and methods

Patients with epilepsy and resting state fMRI were retrospectively identified. Brain network nodes were defined by anatomic parcellation, first in patient space (nodes defined for each patient) and again in template space (same nodes for all patients). Whole-brain weighted graphs were constructed according to pair-wise correlation of BOLD-signal time courses between nodes. The following metrics were then calculated: clustering coefficient, transitivity, modularity, path length, and global efficiency. Metrics computed on graphs in patient space were normalized to the same metric computed on a random network of identical size. A machine learning algorithm was used to predict patient IQ given access to only the network metrics.

Results

Twenty-seven patients (8–18 years) comprised the final study group. All brain networks demonstrated expected small world properties. Accounting for intrinsic population heterogeneity had a significant effect on prediction accuracy. Specifically, transformation of all patients into a common standard space as well as normalization of metrics to those computed on a random network both substantially outperformed the use of non-normalized metrics.

Conclusion

Normalization contributed significantly to accurate subject-level prediction of cognitive function in children with epilepsy. These findings support the potential for quantitative network approaches to contribute clinically meaningful information in children with neurological disorders.

]]>
<![CDATA[Algorithmic bias amplifies opinion fragmentation and polarization: A bounded confidence model]]> https://www.researchpad.co/product?articleinfo=5c8823d7d5eed0c484639133

The flow of information reaching us via the online media platforms is optimized not by the information content or relevance but by popularity and proximity to the target. This is typically performed in order to maximise platform usage. As a side effect, this introduces an algorithmic bias that is believed to enhance fragmentation and polarization of the societal debate. To study this phenomenon, we modify the well-known continuous opinion dynamics model of bounded confidence in order to account for the algorithmic bias and investigate its consequences. In the simplest version of the original model the pairs of discussion participants are chosen at random and their opinions get closer to each other if they are within a fixed tolerance level. We modify the selection rule of the discussion partners: there is an enhanced probability to choose individuals whose opinions are already close to each other, thus mimicking the behavior of online media which suggest interaction with similar peers. As a result we observe: a) an increased tendency towards opinion fragmentation, which emerges also in conditions where the original model would predict consensus, b) increased polarisation of opinions and c) a dramatic slowing down of the speed at which the convergence at the asymptotic state is reached, which makes the system highly unstable. Fragmentation and polarization are augmented by a fragmented initial population.

]]>
<![CDATA[Efficient neural decoding of self-location with a deep recurrent network]]> https://www.researchpad.co/product?articleinfo=5c70678ed5eed0c4847c7217

Place cells in the mammalian hippocampus signal self-location with sparse spatially stable firing fields. Based on observation of place cell activity it is possible to accurately decode an animal’s location. The precision of this decoding sets a lower bound for the amount of information that the hippocampal population conveys about the location of the animal. In this work we use a novel recurrent neural network (RNN) decoder to infer the location of freely moving rats from single unit hippocampal recordings. RNNs are biologically plausible models of neural circuits that learn to incorporate relevant temporal context without the need to make complicated assumptions about the use of prior information to predict the current state. When decoding animal position from spike counts in 1D and 2D-environments, we show that the RNN consistently outperforms a standard Bayesian approach with either flat priors or with memory. In addition, we also conducted a set of sensitivity analysis on the RNN decoder to determine which neurons and sections of firing fields were the most influential. We found that the application of RNNs to neural data allowed flexible integration of temporal context, yielding improved accuracy relative to the more commonly used Bayesian approaches and opens new avenues for exploration of the neural code.

]]>
<![CDATA[Optimizing predictive performance of criminal recidivism models using registration data with binary and survival outcomes]]> https://www.researchpad.co/product?articleinfo=5c8c193ed5eed0c484b4d25f

In a recidivism prediction context, there is no consensus on which modeling strategy should be followed for obtaining an optimal prediction model. In previous papers, a range of statistical and machine learning techniques were benchmarked on recidivism data with a binary outcome. However, two important tree ensemble methods, namely gradient boosting and random forests were not extensively evaluated. In this paper, we further explore the modeling potential of these techniques in the binary outcome criminal prediction context. Additionally, we explore the predictive potential of classical statistical and machine learning methods for censored time-to-event data. A range of statistical manually specified statistical and (semi-)automatic machine learning models is fitted on Dutch recidivism data, both for the binary outcome case and censored outcome case. To enhance generalizability of results, the same models are applied to two historical American data sets, the North Carolina prison data. For all datasets, (semi-) automatic modeling in the binary case seems to provide no improvement over an appropriately manually specified traditional statistical model. There is however evidence of slightly improved performance of gradient boosting in survival data. Results on the reconviction data from two sources suggest that both statistical and machine learning should be tried out for obtaining an optimal model. Even if a flexible black-box model does not improve upon the predictions of a manually specified model, it can serve as a test whether important interactions are missing or other misspecification of the model are present and can thus provide more security in the modeling process.

]]>