CALL CENTERS (CENTRES) - Faculty of Industrial Engineering and ...
CALL CENTERS (CENTRES) - Faculty of Industrial Engineering and ...
CALL CENTERS (CENTRES) - Faculty of Industrial Engineering and ...
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
over a finite interval, based on the number <strong>of</strong> counts in measurement subintervals. Such a linear<br />
arrival-rate function can serve as a component <strong>of</strong> a piecewise-linear approximation to a general<br />
arrival-rate function. We consider ordinary least squares (OLS), iterative weighted least squares<br />
(IWLS) <strong>and</strong> maximum likelihood (ML), all constrained to yield a nonnegative rate function. We<br />
prove that ML coincides with IWLS. As a reference point, we also consider the theoretically optimal<br />
weighted least squares (TWLS), which is least squares with weights inversely proportional<br />
to the variances (which would not be known with data). Overall, ML performs almost as well as<br />
TWLS. We describe computer simulations conducted to evaluate these estimation procedures.<br />
None <strong>of</strong> the procedures differ greatly when the rate function is not near 0 at either end, but<br />
when the rate function is near 0 at one end, TWLS <strong>and</strong> ML are significantly more effective than<br />
OLS. The number <strong>of</strong> measurement subintervals (with fixed total interval) makes surprisingly<br />
little difference when the rate function is not near 0 at either end. The variances are higher<br />
with only two or three subintervals, but there usually is little benefit from going above ten. In<br />
contrast, more measurement intervals help TWLS <strong>and</strong> ML when the rate function is near 0 at<br />
one end. We derive explicit formulas for the OLS variances <strong>and</strong> the asymptotic TWLS variances<br />
(as the number <strong>of</strong> measurement intervals increases), assuming the nonnegativity constraints are<br />
not violated. These formulas reveal the statistical precision <strong>of</strong> the estimators <strong>and</strong> the influence<br />
<strong>of</strong> the parameters <strong>and</strong> the method. Knowing how the variance depends on the interval length<br />
can help determine how to approximate general arrival-rate functions by piecewise-linear ones.<br />
We also develop statistical tests to determine whether the linear Poisson model is appropriate.<br />
Keywords: Digital simulation, Iterative methods, Least-squares approximations, Maximum likelihood<br />
estimation, Parameter estimation, Piecewise linear techniques, Queueing theory, Stochastic<br />
processes, Telecommunication traffic, Nonhomogeneous Poisson process, Piecewise linear approximation,<br />
Linear arrival-rate function, Ordinary least squares, Iterative weighted least squares,<br />
Computer simulations, Statistical precision, Traffic model<br />
(Appears also in Section IX.)<br />
10. Chlebus, E. Empirical validation <strong>of</strong> call holding time distribution in cellular communications<br />
systems. Teletraffic Contributions for the Information Age. Proceedings <strong>of</strong> the 15th International<br />
Teletraffic Congress, ITC-15. Elsevier, Amsterdam, The Netherl<strong>and</strong>s, 1997, 1179–1188.<br />
Abstract. Various probability distributions are fitted to empirical call holding time data collected<br />
in cellular communications systems. Their parameters are determined through maximum<br />
likelihood estimation. A visual plots examination <strong>of</strong> empirical <strong>and</strong> fitted cumulative distribution<br />
functions enables qualitative comparison. Goodness-<strong>of</strong>-fit techniques based on supremum<br />
<strong>and</strong> quadratic empirical distribution function statistics, namely the Kolmogorov-Smirnov <strong>and</strong><br />
Anderson-Darling tests, respectively are implemented to compare quantitatively the produced<br />
fits.<br />
Keywords: Empirical validation, Call holding time distribution, Cellular communications systems,<br />
Probability distributions, Empirical call holding time data, Maximum likelihood estimation,<br />
Cumulative distribution functions, Goodness-<strong>of</strong>-fit techniques, Supremum empirical distribution<br />
function statistics, Quadratic empirical distribution function statistics, Kolmogorov-<br />
Smirnov test, Anderson-Darling test, Exponential distribution, Gamma distribution, Lognormal<br />
distribution<br />
70