Dummy Regressors Research Articles

Potato Consumption Forecasting Based on a Hybrid Stacked Deep Learning Model

AbstractPotato consumption forecasting is crucial for several stakeholders in the food market. Due to the market flexibility, the farmers can manipulate the volumes planted for a given type of produce to reduce costs and improve revenue. Consequently, it means that establishing optimal inventories or inventory levels is possible and critical in that sense for the sellers to avoid either inadequate inventory or excessive inventories that may lead to wastage. In addition, governments can predict future food deficits and put measures in place to guarantee that they have a steady supply of food some of the time, especially in regions that involve the use of potatoes. Increased potato-eating anticipation has advantages for the sellers and buyers of the potatoes. The experiments of this study employed various machine learning and deep learning (DL) models that comprise stacked long short-term memory (Stacked LSTM), convolutional neural network (CNN), random forest (RF), support vector regressor (SVR), K-nearest neighbour regressor (KNN), bagging regressor (BR), and dummy regressor (DR). During the study, it was discovered that the Stacked LSTM model had superior performance compared to the other models. The Stacked LSTM model achieved a mean squared error (MSE) of 0.0081, a mean absolute error (MAE) of 0.0801, a median absolute error (MedAE) of 0.0755, and a coefficient of determination (R2) value of 98.90%. These results demonstrate that our algorithms can reliably forecast global potato consumption until the year 2030.

Predicting State-Level Firearm Suicide Rates: A Machine Learning Approach Using Public Policy Data

IntroductionOver 40,000 people die by suicide annually in the U.S., and firearms are the most lethal suicide method. There is limited evidence on the effectiveness of many state-level policies on reducing firearm suicide. The objective of this study was to identify public policies that best predict state-level firearm suicide rates. MethodsData from the Centers for Disease Control and Prevention's WONDER system and the State Firearm Law Database, a longitudinal catalog of 134 firearm safety laws, were analyzed. The analysis included 1,450 observations from 50 states spanning 1991-2019. An ElasticNet regression technique was used to analyze the relationship between the policy variables and firearm suicide rates. Nested cross-validation was performed to tune the model hyperparameters. The study data were collected and analyzed in 2023 and 2024. ResultsThe optimized ElasticNet approach had a mean MSE of 2.07, which was superior to non-regularized and dummy regressor models. The most influential policies for predicting the firearm suicide rate on average included laws requiring firearm dealers that sell handguns to have a state license and laws requiring individuals to obtain a permit to purchase a firearm through an approval process that includes law enforcement, among others. ConclusionsOn average, firearm suicide rates were lower in state-years that had each influential policy active. Notably, these analyses were ecological and non-causal. However, this study was able to use a supervised machine learning approach with inherent feature selection and many policy types to make predictions using unseen data (i.e., balancing Lasso and Ridge regularization penalties).

Association between wrist-worn free-living accelerometry and hand grip strength in middle-aged and older adults

IntroductionWrist-worn activity monitors have seen widespread adoption in recent times, particularly in young and sport-oriented cohorts, while their usage among older adults has remained relatively low. The main limitations are in regards to the lack of medical insights that current mainstream activity trackers can provide to older subjects. One of the most important research areas under investigation currently is the possibility of extrapolating clinical information from these wearable devices.MethodsThe research question of this study is understanding whether accelerometry data collected for 7-days in free-living environments using a consumer-based wristband device, in conjunction with data-driven machine learning algorithms, is able to predict hand grip strength and possible conditions categorized by hand grip strength in a general population consisting of middle-aged and older adults.ResultsThe results of the regression analysis reveal that the performance of the developed models is notably superior to a simple mean-predicting dummy regressor. While the improvement in absolute terms may appear modest, the mean absolute error (6.32 kg for males and 4.53 kg for females) falls within the range considered sufficiently accurate for grip strength estimation. The classification models, instead, excel in categorizing individuals as frail/pre-frail, or healthy, depending on the T-score levels applied for frailty/pre-frailty definition. While cut-off values for frailty vary, the results suggest that the models can moderately detect characteristics associated with frailty (AUC-ROC: 0.70 for males, and 0.76 for females) and viably detect characteristics associated with frailty/pre-frailty (AUC-ROC: 0.86 for males, and 0.87 for females).Conclusions The results of this study can enable the adoption of wearable devices as an efficient tool for clinical assessment in older adults with multimorbidities, improving and advancing integrated care, diagnosis and early screening of a number of widespread diseases.

A framework for predicting odor threshold values of perfumes by scientific machine learning and transfer learning

Knowledge of odor thresholds is very important for the perfume industry. Due to the difficulty associated with measuring odor thresholds, empirical models capable of estimating these values can be an invaluable contribution to the field. This work developed a framework based on scientific machine learning strategies. A transfer learning-based strategy was devised, where information from a graph convolutional network predicting semantic odor descriptors was used as input data for the feedforward neural network responsible for estimating odor thresholds for chemical substances based on their molecular structures. The predictive performance of this model was compared to a benchmark odor threshold prediction model based on molecular structures that did not utilize transfer learning. Furthermore, the prediction was compared to a correlation previously proposed in the literature and a dummy regressor. Results demonstrated that the transfer learning-based strategy displayed a better predictive performance, suggesting this technique can be useful for predicting odor thresholds.

Identification and estimation of a heteroskedastic censored regression model with random coefficient dummy endogenous regressors

Censoring and the impact of interventions to differ across treated units are two common phenomena in applied micro-econometrics. This paper studies identification and estimation of a heteroskedastic censored regression model with random coefficient dummy regressors. This approach imposes no parametric distributional assumption on error terms and allows for a purely nonparametric participation decision equation. The resulting estimator is shown to be n-consistent and asymptotically normal. Moreover, our estimation approach can readily extend to a censored regression model with multiple random coefficient dummy endogenous regressors. A Monte Carlo simulation shows that our estimator performs reasonably well for finite samples. An application to evaluate the effect of fertility on female labor supply is provided, and the result indicates that the presence of a third child causes work hours per week to decline by approximately 2.5 h.

Predicting the impact of formation protocols on battery lifetime immediately after manufacturing

Predicting the impact of formation protocols on battery lifetime immediately after manufacturing

16 Forecasting Medical Patient Length of Stay at Presentation in an Emergency Department Using Machine Learning

1. Creating a model to predict LOS at presentation, and accessing its performance. 2. Identifying the powerful predictors to inform future models. We collected consecutive medical patients’ data at presentation in the ED of Far Eastern Memorial Hospital, Taiwan from Feb 1 to Sep 30, 2016. Other variables like ambulance diversion duration and ED census were incorporated. In the 8-month period, 47,807 medical visits were observed. After excluding outliers (eg, high utilizers) and those with missing values, a final dataset of 44,148 patient entries and 109 features was created, and then randomly split into a training set and a test set in a ratio of 7:3. The features comprised 4 categories, namely demographics, chief complaints, vital signs, and registration status. Random Forest, the machine-learning (ML) algorithm, was chosen in terms of its interpretability and non-linearity capturability. The model was trained and tuned on the training set, and then evaluated on the test set in terms of mean squared error (MSE). Among the 44,148 observations, the median and the mean of ED LOS measured in hours are 2.30 and 7.73. Logarithmic transformation (2 based) was applied to this outcome variable so that the distribution of the newly derived variable (denoted as ‘EDLOS_hr_log’) was closer to a normal distribution. Table 1 shows the descriptive statistics regarding demographic data in the training set and the test set. Our model scored 2.22 and 2.26 of MSE on the training set and the test set respectively, which signaled the model associated with low variance but high bias. Since there was no benchmark model available for comparison, the concept of a dummy regressor, using the mean of EDLOS_hr_log in the training set as the predicted value for any future cases, was introduced to inform the model’s performance. This dummy regressor scored 3.19 of MSE on the test set; therefore, our model improved the predictions by 29.2%. Moreover, our model could capture the pattern of ED load, described as the moving average of EDLOS_hr_log against date (Figure 1). Figure 2 summarized the top 15 most influential predictors and their categories, which shed a light on the future modeling. The order in this list represents their rankings. The higher it is, the more powerful it is. The intersection of ML and medicine is emerging and evolving. Our model serves as an excellent example how machines can support medical/administrative decisionmaking. We find evidence that ED LOS was predictive at presentation. More importantly, it implies that strategies to tackle ED crowding and reduce ED LOS should incorporate hospital systemic constraints rather than focus only on an ED itself.Table 1Descriptive demographic statistics on the 2 subsets.The training setThe test setPatient Counts30,90313,245Age†51.0 [19.9]51.2 [19.9]Triage 11,120 (3.62%)466 (3.52%)Triage 25,444 (17.6%)2,401 (18.1%)Triage 320,462 (66.2%)8,712 (65.8%)Triage 43,842 (12.4%)1,643 (12.4%)Triage 535 (0.11%)23 (0.17%)EDLOS_hr_log†1.62 [1.77]1.63 [1.79] Open table in a new tab View Large Image Figure ViewerDownload Hi-res image Download (PPT)

Identification and estimation of heteroscedastic binary choice models with endogenous dummy regressors

In this paper, we consider the semiparametric identification and estimation of a heteroscedastic binary choice model with endogenous dummy regressors and no parametric restriction on the distribution of the error term. Our approach addresses various drawbacks associated with previous estimators proposed for this model. It allows for: general multiplicative heteroscedasticity in both selection and outcome equations; a nonparametric selection mechanism; and multiple discrete endogenous regressors. The resulting three‐stage estimator is shown to be asymptotically normal, with a convergence rate that can be arbitrarily close to n−1/2 if certain smoothness assumptions are satisfied. Simulation results show that our estimator performs reasonably well in finite samples. Our approach is then used to study the intergenerational transmission of smoking habits in British households.

Use of crop condition based dummy regressor and weather input for parameter estimation of mustard yield forecast models

Parameter estimation in statistical modelling plays a crucial role in the real world phenomena. Several alternative analyses may be required for the purpose. In this paper, standard linear regression and multivariate statistical analyses were carried out to achieve the district-level rapeseed-mustard yield estimation in Haryana State (India). The study revealed that the zonal weather models incorporating crop condition term as dummy regressor(s) had the desired predictive accuracy. The model based mustard yield(s) indicated good agreement with State Department of Agriculture (DOA) yield estimates by showing 5-10 percent deviations in most of the mustard growing districts however for two-three districts, it gave 12-13 percent deviations possibly due to the smaller set of data available for those districts. The crop yield estimates on the basis of developed models may be obtained 4-5 weeks in advance of the harvest time.

A note on the identification in two equations probit model with dummy endogenous regressor

This paper deals with the question whether exclusion restrictions on the exogenous regressors are necessary to identify two equation probit models with endogenous dummy regressor. We show that Wilde (2000)’s criterion is insufficient for (point) identification.

A Note on the Identification in Two Equations: Probit Model with Dummy Endogenous Regressor

This paper deals with the question whether exclusion restrictions on the exogenous regressors are necessary to identify two equation probit models with endogenous dummy regressor. We show that Wilde (2000)’s criterion is insufficient for (point) identification.

Estimation of a heteroscedastic binary choice model with an endogenous dummy regressor

Estimating binary choice models with endogeneity is of considerable importance in microeconometrics. The leading control function approach does not apply when the endogenous variable is binary. We propose a multi-stage estimation procedure for a heteroscedastic binary choice model with an endogenous dummy under a joint conditional symmetry restriction, which allows us to overcome several drawbacks associated with the existing estimators.

A simple clinical model for planning transfusion quantities in heart surgery.

BackgroundPatients undergoing heart surgery continue to be the largest demand on blood transfusions. The need for transfusion is based on the risk of complications due to poor cell oxygenation, however large transfusions are associated with increased morbidity and risk of mortality in heart surgery patients. The aim of this study was to identify preoperative and intraoperative risk factors for transfusion and create a reliable model for planning transfusion quantities in heart surgery procedures.MethodsWe performed an observational study on 3315 consecutive patients who underwent cardiac surgery between January 2000 and December 2007. To estimate the number of packs of red blood cells (PRBC) transfused during heart surgery, we developed a multivariate regression model with discrete coefficients by selecting dummy variables as regressors in a stepwise manner. Model performance was assessed statistically by splitting cases into training and testing sets of the same size, and clinically by investigating the clinical course details of about one quarter of the patients in whom the difference between model estimates and actual number of PRBC transfused was higher than the root mean squared error.ResultsTen preoperative and intraoperative dichotomous variables were entered in the model. Approximating the regression coefficients to the nearest half unit, each dummy regressor equal to one gave a number of half PRBC. The model assigned 4 units for kidney failure requiring preoperative dialysis, 2.5 units for cardiogenic shock, 2 units for minimum hematocrit at cardiopulmonary bypass less than or equal to 20%, 1.5 units for emergency operation, 1 unit for preoperative hematocrit less than or equal to 40%, cardiopulmonary bypass time greater than 130 minutes and type of surgery different from isolated artery bypass grafting, and 0.5 units for urgent operation, age over 70 years and systemic arterial hypertension.ConclusionsThe regression model proved reliable for quantitative planning of number of PRBC in patients undergoing heart surgery. Besides enabling more rational resource allocation of costly blood-conservation strategies and blood bank resources, the results indicated a strong association between some essential postoperative variables and differences between the model estimate and the actual number of packs transfused.

Estimating multiple equation hybrid models with endogenous dummy regressors*

Estimating multiple equation hybrid models with endogenous dummy regressors is burdensome if the number of dummies and equations is large. Therefore, after clarifying model structure and identification issues the paper presents a generalization of the Amemiya principle. The new estimator reduces the computational costs and is flexible with regard to different types of variables. Furthermore, it is consistent and asymptotically normal distributed.

Simultaneity in multiple equation hybrid models with endogenous dummy regressors

Simultaneity in multiple equation hybrid models with endogenous dummy regressors

Identification of multiple equation probit models with endogenous dummy regressors

This article deals with the question whether exclusion restrictions on the exogenous regressors are necessary to identify multiple equation probit models with endogenous dummy regressors. The contradictory opinions in the literature are discussed, and a simple criterion of avoiding identification problems is formulated.

Generalized Collinearity Diagnostics

Abstract Working in the context of the linear model y = Xβ + ε, we generalize the concept of variance inflation as a measure of collinearity to a subset of parameters in β (denoted by β 1, with the associated columns of X given by X 1). The essential idea underlying this generalization is to examine the impact on the precision of estimation—in particular, the size of an ellipsoidal joint confidence region for β 1—of less-than-optimal selection of other columns of the design matrix (X 2), treating still other columns (X 0) as unalterable, even hypothetically. In typical applications, X 1 contains a set of dummy regressors coding categories of a qualitative variable or a set of polynomial regressors in a quantitative variable; X 2 contains all other regressors in the model, save the constant, which is in X 0. If σ 2 V denotes the realized variance of , and σ 2 U is the variance associated with an optimal selection of X 2, then the corresponding scaled dispersion ellipsoids to be compared are ℰ v = {x : x′V –1 x ≤ 1} and ℰ U = {x : x′U –1 x ≤ 1}, where ℰ U is contained in ℰ v . The two ellipsoids can be compared by considering the radii of ℰ v relative to ℰ U , obtained through the spectral decomposition of V relative to U. We proceed to explore the geometry of generalized variance inflation, to show the relationship of these measures to correlation-matrix determinants and canonical correlations, to consider X matrices structured by relations of marginality among regressor subspaces, to develop the relationship of generalized variance inflation to hypothesis tests in the multivariate normal linear model, and to present several examples.

Advance Notice and Job Search: The Value of an Early Start

Recent studies of the effect of advance notice on jobless duration following displacement misspecify the relationship between these two variables by treating advance notice as a dummy regressor in conventional survival time models. Job search theory, however, suggests that the major impact of advance notice is not to alter the efficiency of search after job loss, but rather to allow workers to begin search before job loss. The authors develop a sequential-regimes job search model which allows search prior to displacement for workers with advance knowledge of layoff, with different search intensities before and after layoff. Maximum-likelihood estimates of the model are then computed using a pooled data set from the 1984 and 1986 Displaced Worker Surveys. Advance knowledge is found to significantly shorten jobless duration for most labor force groups. Using these estimates, the authors simulate the effect of various levels of advance knowledge on jobless duration. Their results indicate that relatively modest advance notice substantially benefits many displaced workers.