Abstract: This study aims to analyze and explore criminal recidivism with different modeling strategies: one based on an explanation of the phenomenon and another based on a prediction task. We compared three common statistical approaches for modeling recidivism: the logistic regression model, the Cox regression model, and the cure rate model. The parameters of these models were estimated from a Bayesian point of view. Additionally, for prediction purposes, we compared the Cox proportional model, a random survival forest, and a deep neural network. To conduct this study, we used a real dataset that corresponds to a cohort of individuals which consisted of men convicted of sexual crimes against women in 1973 in England and Wales. The results show that the logistic regression model tends to give more precise estimations of the probabilities of recidivism both globally and with the subgroups considered, but at the expense of running a model for each moment of the time that is of interest. The cure rate model with a relatively simple distribution, such as Weibull, provides acceptable estimations, and these tend to be better with longer follow-up periods. The Cox regression model can provide the most biased estimations with certain subgroups. The prediction results show the deep neural network's superiority compared to the Cox proportional model and the random survival forest.

Abstract: Neural networks with random weights have the advantage of fast computational time in both training and testing. However, one of the main challenges of single layer feedforward neural networks is the selection of the optimal number of neurons in the hidden layer, since few/many neurons lead to problems of underfitting/overfitting. Adapting Garson's algorithm, this paper introduces a new efficient and fast non-iterative algorithm for the selection of neurons in the hidden layer for randomization based neural networks. The proposed approach is divided into three steps: (1) train the network with h hidden neurons, (2) apply Garson's algorithm to the matrix of the hidden layer, and (3) perform pruning reducing hidden layer neurons based on the harmonic mean. Our experiments in regression and classification problems confirmed that the combination of the pruning technique with these types of neural networks improved their predictive performance in terms of mean square error and accuracy. Additionally, we tested our proposed pruning method with neural networks trained under sequential learning algorithms, where Random Vector Functional Link obtained, in general, the best predictive performance compared to online sequential versions of extreme learning machines and single hidden layer neural network with random weights. (C) 2018 Elsevier B.V. All rights reserved.

Abstract: In last mile distribution, logistics companies typically arrange and plan their routes based on broad estimates of stop delivery times (i.e., the time spent at each stop to deliver goods to final receivers). If these estimates are not accurate, the level of service is degraded, as the promised time window may not be satisfied. The purpose of this work is to assess the feasibility of machine learning techniques to predict stop delivery times. This is done by testing a wide range of machine learning techniques (including different types of ensembles) to (1) predict the stop delivery time and (2) to determine whether the total stop delivery time will exceed a predefined time threshold (classification approach). For the assessment, all models are trained using information generated from GPS data collected in Medellin, Colombia and compared to hazard duration models. The results are threefold. First, the assessment shows that regression-based machine learning approaches are not better than conventional hazard duration models concerning absolute errors of the prediction of the stop delivery times. Second, when the problem is addressed by a classification scheme in which the prediction is aimed to guide whether a stop time will exceed a predefined time, a basic K-nearest-neighbor model outperforms hazard duration models and other machine learning techniques both in accuracy and F-1 score (harmonic mean between precision and recall). Third, the prediction of the exact duration can be improved by combining the classifiers and prediction models or hazard duration models in a two level scheme (first classification then prediction). However, the improvement depends largely on the correct classification (first level).

Abstract: The performance of ground-source heat pumps (GSHP), often expressed as Power drawn and/or the COP, depends on several operating parameters. Manufacturers usually publish such data in tables for certain discrete values of the operating fluid temperatures and flow rates conditions. In actual applications, such as in dynamic simulations of heat pump system integrated to buildings, there is a need to determine equipment performance under operating conditions other than those listed. This paper describes a simplified methodology for predicting the performance of GSHPs using multiple regression (MR) models as applicable to manufacturer data. We find that fitting second-order MR models with eight statistically significant x-variables from 36 observations appropriately selected in the manufacturer catalogue can predict the system global behavior with good accuracy. For the three studied GSHPs, the external prediction error of the MR models identified following the methodology are 0.2%, 0.9% and 1% for heating capacity (HC) predictions and 2.6%, 4.9% and 3.2% for COP predictions. No correlation is found between residuals and the response, thus validating the models. The operational approach appears to be a reliable tool to be integrated in dynamic simulation codes, as the method is applicable to any GSHP catalogue data. (C) 2016 Elsevier Ltd. All rights reserved.