Predictive power

Résumé

The concept of predictive power, the power of a scientific theory to generate testable predictions, differs from explanatory power and descriptive power (where phenomena that are already known are retrospectively explained or described by a given theory) in that it allows a prospective test of theoretical understanding. Examples A classic example of the predictive power of a theory is the discovery of Neptune as a result of predictions made by mathematicians John Couch Adams and Urbain Le Verrier, based on Newton's theory of gravity. Another example of the predictive power of theories or models is Dmitri Mendeleev's use of his periodic table to predict previously undiscovered chemical elements and their properties. Though largely correct, he misjudged the relative atomic masses of tellurium and iodine. Moreover, Charles Darwin used his knowledge of evolution by natural selection to predict that since a plant (Angraecum sesquipedale) with a long spur in its flowers exists,

Source officielle

À propos de ce résultat

Cette page est générée automatiquement et peut contenir des informations qui ne sont pas correctes, complètes, à jour ou pertinentes par rapport à votre recherche. Il en va de même pour toutes les autres pages de ce site. Veillez à vérifier les informations auprès des sources officielles de l'EPFL.

Publications associées

Chargement

Personnes associées

Chargement

Unités associées

Chargement

Concepts associés

Chargement

Cours associés

Chargement

Séances de cours associées

Chargement

Predicting involuntary hospitalization in psychiatry: A machine learning investigation

Mehdi Gholam

Background Coercion in psychiatry is a controversial issue. Identifying its predictors and their interaction using traditional statistical methods is difficult, given the large number of variables involved. The purpose of this study was to use machine-learning (ML) models to identify socio-demographic, clinical and procedural characteristics that predict the use of compulsory admission on a large sample of psychiatric patients. Methods We retrospectively analyzed the routinely collected data of all psychiatric admissions that occurred between 2013 and 2017 in the canton of Vaud, Switzerland (N = 25,584). The main predictors of involuntary hospitalization were identified using two ML algorithms: Classification and Regression Tree (CART) and Random Forests (RFs). Their predictive power was compared with that obtained through traditional logistic regression. Sensitivity analyses were also performed and missing data were imputed through multiple imputation using chain equations. Results The three models achieved similar predictive balanced accuracy, ranging between 68 and 72%. CART showed the lowest predictive power (68%) but the most parsimonious model, allowing to estimate the probability of being involuntarily admitted with only three checks: aggressive behaviors, who referred the patient to hospital and primary diagnosis. The results of CART and RFs on the imputed data were almost identical to those obtained on the original data, confirming the robustness of our models. Conclusions Identifying predictors of coercion is essential to efficiently target the development of professional training, preventive strategies and alternative interventions. ML methodologies could offer new effective tools to achieve this goal, providing accurate but simple models that could be used in clinical practice.

CAMBRIDGE UNIV PRESS2021

Chargement

, ,

In this paper we first recall the recent result that in deep networks a phase transition, analogous to the jamming transition of granular media, delimits the over- and under-parametrized regimes where fitting can or cannot be achieved. The analysis leading to this result support that for proper initialization and architectures, in the whole over-parametrized regime poor minima of the loss are not encountered during training, because the number of constraints that hinders the dynamics is insufficient to allow for the emergence of stable minima. Next, we study systematically how this transition affects generalization properties of the network (i.e. its predictive power). As we increase the number of parameters of a given model, starting from an under-parametrized network, we observe for gradient descent that the generalization error displays three phases: (i) initial decay, (ii) increase until the transition point?where it displays a cusp?and (iii) slow decay toward an asymptote as the network width diverges. However if early stopping is used, the cusp signaling the jamming transition disappears. Thereby we identify the region where the classical phenomenon of over-fitting takes place as the vicinity of the jamming transition, and the region where the model keeps improving with increasing the number of parameters, thus organizing previous empirical observations made in modern neural networks.

IOP PUBLISHING LTD2019

Chargement

, , , ,

Individuals differ in their physiological responsiveness to stressful challenges, and stress potentiates the development of many diseases. Heart rate variability (HRV), a measure of cardiac vagal break, is emerging as a strong index of physiological stress vulnerability. Thus, it is important to develop tools that identify predictive markers of individual differences in HRV responsiveness without exposing subjects to high stress. Here, using machine learning approaches, we show the strong predictive power of high-dimensional locomotor responses during novelty exploration to predict HRV responsiveness during stress exposure. Locomotor responses are collected in two ecologically valid virtual reality scenarios inspired by the animal literature and stress is elicited and measured in a third threatening virtual scenario. Our model's predictions generalize to other stressful challenges and outperforms other stress prediction instruments, such as anxiety questionnaires. Our study paves the way for the development of behavioral digital phenotyping tools for early detection of stress-vulnerable individuals.

2020

Chargement