Abstract

In the past, the p-center problem had been solved by metaheuristic methods such as Harmony Search, Modified-Artificial Bee Colony, Particle Swarm Optimization (PSO) – Euclidean Distance, and Iterated Local Search with Firefly Algorithm, among others. In this study, a relatively new physics-inspired algorithm, called the Ions Motions Algorithm (IMO), is to be tested on the p-center problem. As of this writing and to the best of the researchers’ abilities, IMO had only been tested on a few problems such as benchmark test functions and protein folding predictions. Testing on p-center benchmark data sets with 7,146 data points, 13,509 data points and 16,862 data points, IMO obtained competitive results with past literature using PSO-based methodology. Initial test runs of IMO on a 29-demand point data (5-center Western Sahara) also revealed that using a higher probability criterion (0.30 over 0.05) for the solid phase of the algorithm generated better results on the average. Furthermore, IMO results in this study obtained small relative errors compared to current state-of-art results even with reduced parameter value setting including number of runs and iterations.

Index Terms

nature-inspired algorithms, location-allocation problems, np-hard problems, p-center problems

Abstract

Government hospitals are mandated not to turn away patients seeking medical attention, especially those with emergency conditions. However, in reality, non-admittance in the emergency department (ED) may happen due to many factors such as understaffing and overcrowding in the emergency department. This study was undertaken to address the problem of overcrowding in the ED by predicting hospital admission at triage. The study setting was at a government hospital ED in Region 10, Mindanao. Data were collected from its administrative logbook record from April 1, 2017 to May 31, 2017. There were 3,016 recorded visits but were trimmed down due to missing entries, illegible handwritings, deaths, transfer, refusal or referral to other hospitals. The proposed model using logistic regression used 2,444 data points which has a 72.5% accuracy. The significant predictors or variables of the model are the following: patient type, age group, clinical department and patient’s acuity category and with admission twice more like under a case in internal medicine or in pediatrics.

Index Terms

emergency department, hospital admission, logistic regression, overcrowding, patient’s acuity, triage

Abstract

This study derives the joint posterior distribution of the parameters of the Asymmetric Power Autoregressive Conditional Heteroscedasticity (A-PARCH(p,q)) model using the Bayesian approach. The Markov Chain Monte Carlo (MCMC) method with Metropolis-Hastings (M-H) algorithm is used in estimating the parameters of the model. The procedure is applied to model the returns of S&P 500 stock market. The best fit model for the S&P 500 stock market returns is ARMA(1,1) – A-PARCH(2,1) model with Student's t-distributed innovations.

Index Terms

A-PARCH model, Bayesian inference, estimation, heteroscedasticity, stock market returns

Abstract

Modified band depth (MBD) is a measure that can be used to order complex data such as high-dimensional data. The ordering can be used as a basis for nonparametric methods such as rank tests, classification, clustering and robust inferential tools. However, computational burden limits its applicability especially when iterative procedures are involved. In this paper, faster and more efficient algorithms, resampling- based MBD and ranking-based MBD, are proposed for this type of data. Remarkable computational gains are demonstrated through simulation studies and application to real datasets as compared to MBD algorithm and more prominent as an efficient tool for high-dimensional data analysis especially when the sample size n and d dimension become large.

Index Terms

High-dimensional, Band Depth, Non-parametric methods, Non-parametric classification and clustering

Abstract

Let X1,X2, ... ,Xn and Y1,Y2, ... , Yn be random samples from two exponential populations, with scale parameters, σ1 and σ2, respectively. This paper considers a two-stage sequential procedure to construct fixed-width confidence intervals In for functions of the exponential scale parameters of the form θ = h(σ1, σ2), where h is a real-valued, three-times continuously differential function defined on R+2. A two-stage sequential procedure is proposed for the estimation of θ through the stopping rules md and Nd defined in equations (3) and (4), respectively. Under the assumption that σ1 > σL and σ2 > σG, where σL, σG > 0 are lower bounds known to the experimenter from past experiences, we have shown that the stopping rule Nd is a good estimate of the optimal sample size n ∗defined in (2). We have shown that the proposed two-stage sequential procedure will eventually stop with probability 1, that is, P(Nd < ∞) = 1. Moreover, we also provide the coverage probability of the interval estimates In guaranteeing asymptotic consistency for the parameter θ. Performances of the proposed two-stage methodology is illustrated via simulation using the R programming language on a parameter of the forms θ = (σ1⁄σ2)^r and θ = |σ1−σ2|^r for r>0. Simulation results show that the proposed two-stage procedure is asymptotically consistent.

Index Terms

two-stage sequential procedure, confidence interval, stopping rule, exponential distribution.

Abstract

Abstract—Computerized mastery testing (CMT) classifies an examinee as either master or non-master which has a termination criterion. This study investigates the application of Sequential Probability Ratio Test (SPRT) and Truncated Sequential Probability Ratio Test (TSPRT) in the computerized mastery testing (CMT) using a polytomous generated responses. Specifically, it applies the nominal response model (NRM) since most of the achievement and assessment tests are based in a multiple-choice item format (MCIT) as such in licensure examinations. Simulation studies are conducted and the Average Test Length (ATL) is obtained, of which ATL is defined as the number of items needed to obtain a decision and is use to evaluate the efficiency of the SPRT and TSPRT in CTM. From the results obtained, it revealed that the sequential tests applied for polytomously generated responses in CMT is efficient, i.e., it obtained the optimum number of items needed to make a decision, which in turn is beneficial both economically and time.

Index Terms

Average Test Length (ATL), Computerized Mastery Testing (CMT), Sequential Probability Ratio Test (SPRT), Nominal Response Model (NRM).

Abstract

In this paper, item pool is generated in R-programming language with 100 maximum items calibrated using 3-parameter logistic model from item response theory, arranged according to Fisher information at specified cutting points θ0=0.5, 0.65 and 0.81. Indifference regions are constructed through fixed δ having values 0.2, 0.3 and 0.4 for hypotheses testing with dichotomous classification (master or non-master) given four set of error rates α and β. During intensive computing, optimum stopping rule for proposed truncated sequential procedure is utilized for the case when n≥50 and n≥30 by reiterating the procedure 5000 times for 5000 simulees to obtain the average test length as performance evaluation tool. Another 25-simulation run for each true ability of examinee from ̶ 3 to +3 having 0.25 increment is done in determining the average sample number to reach a decision plotted via OC and ASN curves. The results of thorough simulations designed in this paper reveals that by truncating the existing procedure results to shorter test length as expected but with higher accuracy and better precision.

Index Terms

CMT, R software, simulation, truncated SPRT

Abstract

The purpose of this study is to investigate the performance of three estimation procedures, namely, the marginal maximum likelihood estimation (MMLE), marginalized Bayesian estimation (MBE), and the Markov chain Monte Carlo (MCMC) in estimating the item parameters of a four parameter logistic (4-PL) item response model. Simulated data are used to obtain item parameter estimates using MMLE, MBE, and MCMC in R language program. Different sample sizes and test lengths are also investigated to reveal various test situations. The results show that the MMLE and MBE methods are affected by varying sample sizes and test lengths, while the MCMC method is affected by the prior specifications. Given that the prior distributions of the item parameters of a 4-PL IRT model are correctly specified, the MCMC method performs best in estimating the item parameters compared with estimates using the MMLE and MBE methods.

Index Terms

4-PL IRT Model, MMLE, MBE, MCMC