Random prism: an alternative to random forests

Ensemble learning techniques generate multiple classifiers, so called base classifiers, whose combined classification results are used in order to increase the overall classification accuracy. In most ensemble classifiers the base classifiers are based on the Top Down Induction of Decision Trees (TDIDT) approach. However, an alternative approach for the induction of rule based classifiers is the Prism family of algorithms. Prism algorithms produce modular classification rules that do not necessarily fit into a decision tree structure. Prism classification rulesets achieve a comparable and sometimes higher classification accuracy compared with decision tree classifiers, if the data is noisy and large. Yet Prism still suffers from overfitting on noisy and large datasets. In practice ensemble techniques tend to reduce the overfitting, however there exists no ensemble learner for modular classification rule inducers such as the Prism family of algorithms. This article describes the first development of an ensemble learner based on the Prism family of algorithms in order to enhance Prism’s classification accuracy by reducing overfitting.

Random Prism: a noise-tolerant alternative to Random Forests

Ensemble learning can be used to increase the overall classification accuracy of a classifier by generating multiple base classifiers and combining their classification results. A frequently used family of base classifiers for ensemble learning are decision trees. However, alternative approaches can potentially be used, such as the Prism family of algorithms that also induces classification rules. Compared with decision trees, Prism algorithms generate modular classification rules that cannot necessarily be represented in the form of a decision tree. Prism algorithms produce a similar classification accuracy compared with decision trees. However, in some cases, for example, if there is noise in the training and test data, Prism algorithms can outperform decision trees by achieving a higher classification accuracy. However, Prism still tends to overfit on noisy data; hence, ensemble learners have been adopted in this work to reduce the overfitting. This paper describes the development of an ensemble learner using a member of the Prism family as the base classifier to reduce the overfitting of Prism algorithms on noisy datasets. The developed ensemble classifier is compared with a stand-alone Prism classifier in terms of classification accuracy and resistance to noise.

Radiation metabolomics. 3. Biomarker discovery in the urine of gamma-irradiated rats using a simplified metabolomics protocol of gas chromatography-mass spectrometry combined with random forests machine learning algorithm

Abstract Radiation metabolomics employing mass spectral technologies represents a plausible means of high-throughput minimally invasive radiation biodosimetry. A simplified metabolomics protocol is described that employs ubiquitous gas chromatography-mass spectrometry and open source software including random forests machine learning algorithm to uncover latent biomarkers of 3 Gy gamma radiation in rats. Urine was collected from six male Wistar rats and six sham-irradiated controls for 7 days, 4 prior to irradiation and 3 after irradiation. Water and food consumption, urine volume, body weight, and sodium, potassium, calcium, chloride, phosphate and urea excretion showed major effects from exposure to gamma radiation. The metabolomics protocol uncovered several urinary metabolites that were significantly up-regulated (glyoxylate, threonate, thymine, uracil, p-cresol) and down-regulated (citrate, 2-oxoglutarate, adipate, pimelate, suberate, azelaate) as a result of radiation exposure. Thymine and uracil were shown to derive largely from thymidine and 2'-deoxyuridine, which are known radiation biomarkers in the mouse. The radiation metabolomic phenotype in rats appeared to derive from oxidative stress and effects on kidney function. Gas chromatography-mass spectrometry is a promising platform on which to develop the field of radiation metabolomics further and to assist in the design of instrumentation for use in detecting biological consequences of environmental radiation release.

An empirical evaluation of the Random Forests classifier models for variable selection in a large-scale lung cancer case-control study

Random Forests™ is reported to be one of the most accurate classification algorithms in complex data analysis. It shows excellent performance even when most predictors are noisy and the number of variables is much larger than the number of observations. In this thesis Random Forests was applied to a large-scale lung cancer case-control study. A novel way of automatically selecting prognostic factors was proposed. Also, synthetic positive control was used to validate Random Forests method. Throughout this study we showed that Random Forests can deal with large number of weak input variables without overfitting. It can account for non-additive interactions between these input variables. Random Forests can also be used for variable selection without being adversely affected by collinearities. ^ Random Forests can deal with the large-scale data sets without rigorous data preprocessing. It has robust variable importance ranking measure. Proposed is a novel variable selection method in context of Random Forests that uses the data noise level as the cut-off value to determine the subset of the important predictors. This new approach enhanced the ability of the Random Forests algorithm to automatically identify important predictors for complex data. The cut-off value can also be adjusted based on the results of the synthetic positive control experiments. ^ When the data set had high variables to observations ratio, Random Forests complemented the established logistic regression. This study suggested that Random Forests is recommended for such high dimensionality data. One can use Random Forests to select the important variables and then use logistic regression or Random Forests itself to estimate the effect size of the predictors and to classify new observations. ^ We also found that the mean decrease of accuracy is a more reliable variable ranking measurement than mean decrease of Gini. ^

Classification of Southern Ocean krill and icefish echoes using random forests

Acknowledgements The authors thank the crews, fishers, and scientists who conducted the various surveys from which data were obtained. This work was supported by the Government of South Georgia and South Sandwich Islands. Additional logistical support provided by The South Atlantic Environmental Research Institute, with thanks to Paul Brickle. PF receives funding from the MASTS pooling initiative (TheMarine Alliance for Science and Technology for Scotland), and their support is gratefully acknowledged. MASTS is funded by the Scottish Funding Council (grant reference HR09011) and contributing institutions. SF is funded by the Natural Environment Research Council, and data were provided from the British Antarctic Survey Ecosystems Long-term Monitoring and Surveys programme as part of the BAS Polar Science for Planet Earth Programme. The authors also thank the anonymous referees for their helpful suggestions on an earlier version of this manuscript.

Machine learning per l'idrologia: applicazione del metodo random forests per la previsione degli eventi di piena fluviale con un approccio probabilistico

Le tecniche di Machine Learning sono molto utili in quanto consento di massimizzare l’utilizzo delle informazioni in tempo reale. Il metodo Random Forests può essere annoverato tra le tecniche di Machine Learning più recenti e performanti. Sfruttando le caratteristiche e le potenzialità di questo metodo, la presente tesi di dottorato affronta due casi di studio differenti; grazie ai quali è stato possibile elaborare due differenti modelli previsionali. Il primo caso di studio si è incentrato sui principali fiumi della regione Emilia-Romagna, caratterizzati da tempi di risposta molto brevi. La scelta di questi fiumi non è stata casuale: negli ultimi anni, infatti, in detti bacini si sono verificati diversi eventi di piena, in gran parte di tipo “flash flood”. Il secondo caso di studio riguarda le sezioni principali del fiume Po, dove il tempo di propagazione dell’onda di piena è maggiore rispetto ai corsi d’acqua del primo caso di studio analizzato. Partendo da una grande quantità di dati, il primo passo è stato selezionare e definire i dati in ingresso in funzione degli obiettivi da raggiungere, per entrambi i casi studio. Per l’elaborazione del modello relativo ai fiumi dell’Emilia-Romagna, sono stati presi in considerazione esclusivamente i dati osservati; a differenza del bacino del fiume Po in cui ai dati osservati sono stati affiancati anche i dati di previsione provenienti dalla catena modellistica Mike11 NAM/HD. Sfruttando una delle principali caratteristiche del metodo Random Forests, è stata stimata una probabilità di accadimento: questo aspetto è fondamentale sia nella fase tecnica che in fase decisionale per qualsiasi attività di intervento di protezione civile. L'elaborazione dei dati e i dati sviluppati sono stati effettuati in ambiente R. Al termine della fase di validazione, gli incoraggianti risultati ottenuti hanno permesso di inserire il modello sviluppato nel primo caso studio all’interno dell’architettura operativa di FEWS.

A scalable expressive ensemble learning using Random Prism: a MapReduce approach

The induction of classification rules from previously unseen examples is one of the most important data mining tasks in science as well as commercial applications. In order to reduce the influence of noise in the data, ensemble learners are often applied. However, most ensemble learners are based on decision tree classifiers which are affected by noise. The Random Prism classifier has recently been proposed as an alternative to the popular Random Forests classifier, which is based on decision trees. Random Prism is based on the Prism family of algorithms, which is more robust to noise. However, like most ensemble classification approaches, Random Prism also does not scale well on large training data. This paper presents a thorough discussion of Random Prism and a recently proposed parallel version of it called Parallel Random Prism. Parallel Random Prism is based on the MapReduce programming paradigm. The paper provides, for the first time, novel theoretical analysis of the proposed technique and in-depth experimental study that show that Parallel Random Prism scales well on a large number of training examples, a large number of data features and a large number of processors. Expressiveness of decision rules that our technique produces makes it a natural choice for Big Data applications where informed decision making increases the user’s trust in the system.

QSAR modeling of antitubercular activity of diverse organic compounds

Tuberculosis (TB) is a worldwide infectious disease that has shown over time extremely high mortality levels. The urgent need to develop new antitubercular drugs is due to the increasing rate of appearance of multi-drug resistant strains to the commonly used drugs, and the longer durations of therapy and recovery, particularly in immuno-compromised patients. The major goal of the present study is the exploration of data from different families of compounds through the use of a variety of machine learning techniques so that robust QSAR-based models can be developed to further guide in the quest for new potent anti-TB compounds. Eight QSAR models were built using various types of descriptors (from ADRIANA.Code and Dragon software) with two publicly available structurally diverse data sets, including recent data deposited in PubChem. QSAR methodologies used Random Forests and Associative Neural Networks. Predictions for the external evaluation sets obtained accuracies in the range of 0.76-0.88 (for active/inactive classifications) and Q(2)=0.66-0.89 for regressions. Models developed in this study can be used to estimate the anti-TB activity of drug candidates at early stages of drug development (C) 2011 Elsevier B.V. All rights reserved.

Automatic learning for the classification of chemical reactions and in statistical thermodynamics

This Thesis describes the application of automatic learning methods for a) the classification of organic and metabolic reactions, and b) the mapping of Potential Energy Surfaces(PES). The classification of reactions was approached with two distinct methodologies: a representation of chemical reactions based on NMR data, and a representation of chemical reactions from the reaction equation based on the physico-chemical and topological features of chemical bonds. NMR-based classification of photochemical and enzymatic reactions. Photochemical and metabolic reactions were classified by Kohonen Self-Organizing Maps (Kohonen SOMs) and Random Forests (RFs) taking as input the difference between the 1H NMR spectra of the products and the reactants. The development of such a representation can be applied in automatic analysis of changes in the 1H NMR spectrum of a mixture and their interpretation in terms of the chemical reactions taking place. Examples of possible applications are the monitoring of reaction processes, evaluation of the stability of chemicals, or even the interpretation of metabonomic data. A Kohonen SOM trained with a data set of metabolic reactions catalysed by transferases was able to correctly classify 75% of an independent test set in terms of the EC number subclass. Random Forests improved the correct predictions to 79%. With photochemical reactions classified into 7 groups, an independent test set was classified with 86-93% accuracy. The data set of photochemical reactions was also used to simulate mixtures with two reactions occurring simultaneously. Kohonen SOMs and Feed-Forward Neural Networks (FFNNs) were trained to classify the reactions occurring in a mixture based on the 1H NMR spectra of the products and reactants. Kohonen SOMs allowed the correct assignment of 53-63% of the mixtures (in a test set). Counter-Propagation Neural Networks (CPNNs) gave origin to similar results. The use of supervised learning techniques allowed an improvement in the results. They were improved to 77% of correct assignments when an ensemble of ten FFNNs were used and to 80% when Random Forests were used. This study was performed with NMR data simulated from the molecular structure by the SPINUS program. In the design of one test set, simulated data was combined with experimental data. The results support the proposal of linking databases of chemical reactions to experimental or simulated NMR data for automatic classification of reactions and mixtures of reactions. Genome-scale classification of enzymatic reactions from their reaction equation. The MOLMAP descriptor relies on a Kohonen SOM that defines types of bonds on the basis of their physico-chemical and topological properties. The MOLMAP descriptor of a molecule represents the types of bonds available in that molecule. The MOLMAP descriptor of a reaction is defined as the difference between the MOLMAPs of the products and the reactants, and numerically encodes the pattern of bonds that are broken, changed, and made during a chemical reaction. The automatic perception of chemical similarities between metabolic reactions is required for a variety of applications ranging from the computer validation of classification systems, genome-scale reconstruction (or comparison) of metabolic pathways, to the classification of enzymatic mechanisms. Catalytic functions of proteins are generally described by the EC numbers that are simultaneously employed as identifiers of reactions, enzymes, and enzyme genes, thus linking metabolic and genomic information. Different methods should be available to automatically compare metabolic reactions and for the automatic assignment of EC numbers to reactions still not officially classified. In this study, the genome-scale data set of enzymatic reactions available in the KEGG database was encoded by the MOLMAP descriptors, and was submitted to Kohonen SOMs to compare the resulting map with the official EC number classification, to explore the possibility of predicting EC numbers from the reaction equation, and to assess the internal consistency of the EC classification at the class level. A general agreement with the EC classification was observed, i.e. a relationship between the similarity of MOLMAPs and the similarity of EC numbers. At the same time, MOLMAPs were able to discriminate between EC sub-subclasses. EC numbers could be assigned at the class, subclass, and sub-subclass levels with accuracies up to 92%, 80%, and 70% for independent test sets. The correspondence between chemical similarity of metabolic reactions and their MOLMAP descriptors was applied to the identification of a number of reactions mapped into the same neuron but belonging to different EC classes, which demonstrated the ability of the MOLMAP/SOM approach to verify the internal consistency of classifications in databases of metabolic reactions. RFs were also used to assign the four levels of the EC hierarchy from the reaction equation. EC numbers were correctly assigned in 95%, 90%, 85% and 86% of the cases (for independent test sets) at the class, subclass, sub-subclass and full EC number level,respectively. Experiments for the classification of reactions from the main reactants and products were performed with RFs - EC numbers were assigned at the class, subclass and sub-subclass level with accuracies of 78%, 74% and 63%, respectively. In the course of the experiments with metabolic reactions we suggested that the MOLMAP / SOM concept could be extended to the representation of other levels of metabolic information such as metabolic pathways. Following the MOLMAP idea, the pattern of neurons activated by the reactions of a metabolic pathway is a representation of the reactions involved in that pathway - a descriptor of the metabolic pathway. This reasoning enabled the comparison of different pathways, the automatic classification of pathways, and a classification of organisms based on their biochemical machinery. The three levels of classification (from bonds to metabolic pathways) allowed to map and perceive chemical similarities between metabolic pathways even for pathways of different types of metabolism and pathways that do not share similarities in terms of EC numbers. Mapping of PES by neural networks (NNs). In a first series of experiments, ensembles of Feed-Forward NNs (EnsFFNNs) and Associative Neural Networks (ASNNs) were trained to reproduce PES represented by the Lennard-Jones (LJ) analytical potential function. The accuracy of the method was assessed by comparing the results of molecular dynamics simulations (thermal, structural, and dynamic properties) obtained from the NNs-PES and from the LJ function. The results indicated that for LJ-type potentials, NNs can be trained to generate accurate PES to be used in molecular simulations. EnsFFNNs and ASNNs gave better results than single FFNNs. A remarkable ability of the NNs models to interpolate between distant curves and accurately reproduce potentials to be used in molecular simulations is shown. The purpose of the first study was to systematically analyse the accuracy of different NNs. Our main motivation, however, is reflected in the next study: the mapping of multidimensional PES by NNs to simulate, by Molecular Dynamics or Monte Carlo, the adsorption and self-assembly of solvated organic molecules on noble-metal electrodes. Indeed, for such complex and heterogeneous systems the development of suitable analytical functions that fit quantum mechanical interaction energies is a non-trivial or even impossible task. The data consisted of energy values, from Density Functional Theory (DFT) calculations, at different distances, for several molecular orientations and three electrode adsorption sites. The results indicate that NNs require a data set large enough to cover well the diversity of possible interaction sites, distances, and orientations. NNs trained with such data sets can perform equally well or even better than analytical functions. Therefore, they can be used in molecular simulations, particularly for the ethanol/Au (111) interface which is the case studied in the present Thesis. Once properly trained, the networks are able to produce, as output, any required number of energy points for accurate interpolations.

Using data mining techniques to support breast cancer diagnosis

More than ever, there is an increase of the number of decision support methods and computer aided diagnostic systems applied to various areas of medicine. In breast cancer research, many works have been done in order to reduce false-positives when used as a double reading method. In this study, we aimed to present a set of data mining techniques that were applied to approach a decision support system in the area of breast cancer diagnosis. This method is geared to assist clinical practice in identifying mammographic findings such as microcalcifications, masses and even normal tissues, in order to avoid misdiagnosis. In this work a reliable database was used, with 410 images from about 115 patients, containing previous reviews performed by radiologists as microcalcifications, masses and also normal tissue findings. Throughout this work, two feature extraction techniques were used: the gray level co-occurrence matrix and the gray level run length matrix. For classification purposes, we considered various scenarios according to different distinct patterns of injuries and several classifiers in order to distinguish the best performance in each case described. The many classifiers used were Naïve Bayes, Support Vector Machines, k-nearest Neighbors and Decision Trees (J48 and Random Forests). The results in distinguishing mammographic findings revealed great percentages of PPV and very good accuracy values. Furthermore, it also presented other related results of classification of breast density and BI-RADS® scale. The best predictive method found for all tested groups was the Random Forest classifier, and the best performance has been achieved through the distinction of microcalcifications. The conclusions based on the several tested scenarios represent a new perspective in breast cancer diagnosis using data mining techniques.

Análise comparada da segmentação e classificação orientada por objectos de uma imagem de alta resolução WORLDVIEW-2

Dissertação de Mestrado em Gestão do Território, Especialização em Detecção Remota e Sistemas de Informação Geográfica

A behavioral ecology of shermen: hidden stories from trajectory data in the Northern Humboldt Current System

This work proposes an original contribution to the understanding of shermen spatial behavior, based on the behavioral ecology and movement ecology paradigms. Through the analysis of Vessel Monitoring System (VMS) data, we characterized the spatial behavior of Peruvian anchovy shermen at di erent scales: (1) the behavioral modes within shing trips (i.e., searching, shing and cruising); (2) the behavioral patterns among shing trips; (3) the behavioral patterns by shing season conditioned by ecosystem scenarios; and (4) the computation of maps of anchovy presence proxy from the spatial patterns of behavioral mode positions. At the rst scale considered, we compared several Markovian (hidden Markov and semi-Markov models) and discriminative models (random forests, support vector machines and arti cial neural networks) for inferring the behavioral modes associated with VMS tracks. The models were trained under a supervised setting and validated using tracks for which behavioral modes were known (from on-board observers records). Hidden semi-Markov models performed better, and were retained for inferring the behavioral modes on the entire VMS dataset. At the second scale considered, each shing trip was characterized by several features, including the time spent within each behavioral mode. Using a clustering analysis, shing trip patterns were classi ed into groups associated to management zones, eet segments and skippers' personalities. At the third scale considered, we analyzed how ecological conditions shaped shermen behavior. By means of co-inertia analyses, we found signi cant associations between shermen, anchovy and environmental spatial dynamics, and shermen behavioral responses were characterized according to contrasted environmental scenarios. At the fourth scale considered, we investigated whether the spatial behavior of shermen re ected to some extent the spatial distribution of anchovy. Finally, this work provides a wider view of shermen behavior: shermen are not only economic agents, but they are also foragers, constrained by ecosystem variability. To conclude, we discuss how these ndings may be of importance for sheries management, collective behavior analyses and end-to-end models.

Batch effect confounding leads to strong bias in performance estimates obtained by cross-validation.

BACKGROUND: With the large amount of biological data that is currently publicly available, many investigators combine multiple data sets to increase the sample size and potentially also the power of their analyses. However, technical differences ("batch effects") as well as differences in sample composition between the data sets may significantly affect the ability to draw generalizable conclusions from such studies. FOCUS: The current study focuses on the construction of classifiers, and the use of cross-validation to estimate their performance. In particular, we investigate the impact of batch effects and differences in sample composition between batches on the accuracy of the classification performance estimate obtained via cross-validation. The focus on estimation bias is a main difference compared to previous studies, which have mostly focused on the predictive performance and how it relates to the presence of batch effects. DATA: We work on simulated data sets. To have realistic intensity distributions, we use real gene expression data as the basis for our simulation. Random samples from this expression matrix are selected and assigned to group 1 (e.g., 'control') or group 2 (e.g., 'treated'). We introduce batch effects and select some features to be differentially expressed between the two groups. We consider several scenarios for our study, most importantly different levels of confounding between groups and batch effects. METHODS: We focus on well-known classifiers: logistic regression, Support Vector Machines (SVM), k-nearest neighbors (kNN) and Random Forests (RF). Feature selection is performed with the Wilcoxon test or the lasso. Parameter tuning and feature selection, as well as the estimation of the prediction performance of each classifier, is performed within a nested cross-validation scheme. The estimated classification performance is then compared to what is obtained when applying the classifier to independent data.

Community Policing in Switzerland's Major Urban Areas: An Observational Study of the Implementation and Impact Using Geospatial Data Mining

La présente étude est à la fois une évaluation du processus de la mise en oeuvre et des impacts de la police de proximité dans les cinq plus grandes zones urbaines de Suisse - Bâle, Berne, Genève, Lausanne et Zurich. La police de proximité (community policing) est à la fois une philosophie et une stratégie organisationnelle qui favorise un partenariat renouvelé entre la police et les communautés locales dans le but de résoudre les problèmes relatifs à la sécurité et à l'ordre public. L'évaluation de processus a analysé des données relatives aux réformes internes de la police qui ont été obtenues par l'intermédiaire d'entretiens semi-structurés avec des administrateurs clés des cinq départements de police, ainsi que dans des documents écrits de la police et d'autres sources publiques. L'évaluation des impacts, quant à elle, s'est basée sur des variables contextuelles telles que des statistiques policières et des données de recensement, ainsi que sur des indicateurs d'impacts construit à partir des données du Swiss Crime Survey (SCS) relatives au sentiment d'insécurité, à la perception du désordre public et à la satisfaction de la population à l'égard de la police. Le SCS est un sondage régulier qui a permis d'interroger des habitants des cinq grandes zones urbaines à plusieurs reprises depuis le milieu des années 1980. L'évaluation de processus a abouti à un « Calendrier des activités » visant à créer des données de panel permettant de mesurer les progrès réalisés dans la mise en oeuvre de la police de proximité à l'aide d'une grille d'évaluation à six dimensions à des intervalles de cinq ans entre 1990 et 2010. L'évaluation des impacts, effectuée ex post facto, a utilisé un concept de recherche non-expérimental (observational design) dans le but d'analyser les impacts de différents modèles de police de proximité dans des zones comparables à travers les cinq villes étudiées. Les quartiers urbains, délimités par zone de code postal, ont ainsi été regroupés par l'intermédiaire d'une typologie réalisée à l'aide d'algorithmes d'apprentissage automatique (machine learning). Des algorithmes supervisés et non supervisés ont été utilisés sur les données à haute dimensionnalité relatives à la criminalité, à la structure socio-économique et démographique et au cadre bâti dans le but de regrouper les quartiers urbains les plus similaires dans des clusters. D'abord, les cartes auto-organisatrices (self-organizing maps) ont été utilisées dans le but de réduire la variance intra-cluster des variables contextuelles et de maximiser simultanément la variance inter-cluster des réponses au sondage. Ensuite, l'algorithme des forêts d'arbres décisionnels (random forests) a permis à la fois d'évaluer la pertinence de la typologie de quartier élaborée et de sélectionner les variables contextuelles clés afin de construire un modèle parcimonieux faisant un minimum d'erreurs de classification. Enfin, pour l'analyse des impacts, la méthode des appariements des coefficients de propension (propensity score matching) a été utilisée pour équilibrer les échantillons prétest-posttest en termes d'âge, de sexe et de niveau d'éducation des répondants au sein de chaque type de quartier ainsi identifié dans chacune des villes, avant d'effectuer un test statistique de la différence observée dans les indicateurs d'impacts. De plus, tous les résultats statistiquement significatifs ont été soumis à une analyse de sensibilité (sensitivity analysis) afin d'évaluer leur robustesse face à un biais potentiel dû à des covariables non observées. L'étude relève qu'au cours des quinze dernières années, les cinq services de police ont entamé des réformes majeures de leur organisation ainsi que de leurs stratégies opérationnelles et qu'ils ont noué des partenariats stratégiques afin de mettre en oeuvre la police de proximité. La typologie de quartier développée a abouti à une réduction de la variance intra-cluster des variables contextuelles et permet d'expliquer une partie significative de la variance inter-cluster des indicateurs d'impacts avant la mise en oeuvre du traitement. Ceci semble suggérer que les méthodes de géocomputation aident à équilibrer les covariables observées et donc à réduire les menaces relatives à la validité interne d'un concept de recherche non-expérimental. Enfin, l'analyse des impacts a révélé que le sentiment d'insécurité a diminué de manière significative pendant la période 2000-2005 dans les quartiers se trouvant à l'intérieur et autour des centres-villes de Berne et de Zurich. Ces améliorations sont assez robustes face à des biais dus à des covariables inobservées et covarient dans le temps et l'espace avec la mise en oeuvre de la police de proximité. L'hypothèse alternative envisageant que les diminutions observées dans le sentiment d'insécurité soient, partiellement, un résultat des interventions policières de proximité semble donc être aussi plausible que l'hypothèse nulle considérant l'absence absolue d'effet. Ceci, même si le concept de recherche non-expérimental mis en oeuvre ne peut pas complètement exclure la sélection et la régression à la moyenne comme explications alternatives. The current research project is both a process and impact evaluation of community policing in Switzerland's five major urban areas - Basel, Bern, Geneva, Lausanne, and Zurich. Community policing is both a philosophy and an organizational strategy that promotes a renewed partnership between the police and the community to solve problems of crime and disorder. The process evaluation data on police internal reforms were obtained through semi-structured interviews with key administrators from the five police departments as well as from police internal documents and additional public sources. The impact evaluation uses official crime records and census statistics as contextual variables as well as Swiss Crime Survey (SCS) data on fear of crime, perceptions of disorder, and public attitudes towards the police as outcome measures. The SCS is a standing survey instrument that has polled residents of the five urban areas repeatedly since the mid-1980s. The process evaluation produced a "Calendar of Action" to create panel data to measure community policing implementation progress over six evaluative dimensions in intervals of five years between 1990 and 2010. The impact evaluation, carried out ex post facto, uses an observational design that analyzes the impact of the different community policing models between matched comparison areas across the five cities. Using ZIP code districts as proxies for urban neighborhoods, geospatial data mining algorithms serve to develop a neighborhood typology in order to match the comparison areas. To this end, both unsupervised and supervised algorithms are used to analyze high-dimensional data on crime, the socio-economic and demographic structure, and the built environment in order to classify urban neighborhoods into clusters of similar type. In a first step, self-organizing maps serve as tools to develop a clustering algorithm that reduces the within-cluster variance in the contextual variables and simultaneously maximizes the between-cluster variance in survey responses. The random forests algorithm then serves to assess the appropriateness of the resulting neighborhood typology and to select the key contextual variables in order to build a parsimonious model that makes a minimum of classification errors. Finally, for the impact analysis, propensity score matching methods are used to match the survey respondents of the pretest and posttest samples on age, gender, and their level of education for each neighborhood type identified within each city, before conducting a statistical test of the observed difference in the outcome measures. Moreover, all significant results were subjected to a sensitivity analysis to assess the robustness of these findings in the face of potential bias due to some unobserved covariates. The study finds that over the last fifteen years, all five police departments have undertaken major reforms of their internal organization and operating strategies and forged strategic partnerships in order to implement community policing. The resulting neighborhood typology reduced the within-cluster variance of the contextual variables and accounted for a significant share of the between-cluster variance in the outcome measures prior to treatment, suggesting that geocomputational methods help to balance the observed covariates and hence to reduce threats to the internal validity of an observational design. Finally, the impact analysis revealed that fear of crime dropped significantly over the 2000-2005 period in the neighborhoods in and around the urban centers of Bern and Zurich. These improvements are fairly robust in the face of bias due to some unobserved covariate and covary temporally and spatially with the implementation of community policing. The alternative hypothesis that the observed reductions in fear of crime were at least in part a result of community policing interventions thus appears at least as plausible as the null hypothesis of absolutely no effect, even if the observational design cannot completely rule out selection and regression to the mean as alternative explanations.