"... In PAGE 5: ... TABLE1 NN OPTIMAL SETTINGS FOR 5- AND 10-MINUTE PREDICTIONS BASED ON AARE .... In PAGE 56: ... 25 identifies the optimal setting as Jordan-Elman NN with inputs from upstream (Z) and current station(Y), downstream (X) and LTM input. TABLE1 TABLE 4 Step 4: Next, the CBR prediction is evaluated. For each prediction horizon, the same testing set used for evaluating the NNs optimal settings in Step3 is used for CBR to generate the probability of occurrence plots in FIGURE 12 - FIGURE 15.... In PAGE 60: ... The tables were sorted by the computed performance measure for the hybrid NN-CBR approach. Considering the performance measures for the NN approach only, the TABLE1 through also show that no particular network topology seemed to have outperformed the others for all cases. The same applies to the type of inputs and the inclusion of LTM component.... ..."

"... In PAGE 8: ... We also computed a residual error value, as the difference between the actual and predicted laten- cies. Table4 shows the results from this analysis, using... In PAGE 8: ...Table 4: Quality of predictions based on equation 1 In Table4 , the median residuals are always negative, im- plying that equation 1 overestimates the transfer latency more often than it underestimates it. However, the mean residuals are always positive, because the equation apos;s un- derestimates are more wrong (in absolute terms) than its overestimates.... ..."

"... In PAGE 3: ... For example, if all three neighboring macroblocks have been intra coded, the target macroblock likely belongs to an area with non- motion changes. Therefore, our nearest-neighbors search algorithm employs the prediction method illustrated in Table1 . The table shows that the predicted motion vector is derived from previously coded motion vectors of inter coded macroblocks of the macroblocks shown in Figure 2.... In PAGE 4: ... Thus, the search path consists of an orderly sequence of search centers. Such a path is abandoned when motion vector prediction is very unreliable as indicated in Table1 , and any path with a more extensive range can instead be employed. In this work, the well-known three-step search path appears to be a good alternative.... ..."

"... In PAGE 5: ... The results can be conveniently summarized by the number of residues in each prediction class (Table 2). From the raw prediction data in Table2 it is possible to derive all the different performance measures discussed above (Table 3). The order of the algorithms in these tables is determined by the probabilityexcess, ourfavoredperformancemeasure.... ..."

"... In PAGE 5: ...N AARE ............................................................................................................................ 53 TABLE2 NN OPTIMAL SETTINGS FOR 15- AND 20-MINUTE PREDICTIONS BASED ON AARE .... In PAGE 56: ... The network setting with the minimum error is selected as the one that should be used further in the prediction for a certain traffic conditions combination. For example, if one wants to estimate the speed in 10-minute prediction horizon, at free-flow prevailing traffic conditions at each of the three stations, and if AARE is considered as performance measure, than in TABLE2 d) case no. 25 identifies the optimal setting as Jordan-Elman NN with inputs from upstream (Z) and current station(Y), downstream (X) and LTM input.... ..."

"... In PAGE 20: ... Duke and Mirant do not own units located in the SCAQMD region. Table1 reports the results from estimating the following regression (3) Consistent with our hypothesis, the estimates of (00 , (01 and 800 and 801 are all positive, and all but the estimate of (00 are statistically significantly different from zero. Moreover, we find that the joint null hypothesis H: $1 = $2 = $3 = $4 = 0 cannot be rejected.... In PAGE 21: ... Table 2 reports the results of this regression. Although the transactions year indicator variables for 2000 and 2001 are estimated to be very large and positive, the estimates of (00 , (01 and 800 and 801 are all positive, different from Table1 , only 800 and 801 are statistically significantly different from zero. The joint null hypothesis H: $1 = $2 = $3 = $4 = 0 still cannot be rejected.... ..."

"... In PAGE 11: ... Once this calculation has been performed for all symbol pairs, those with greatest savings are turned into generalisations. Table2 shows the top ranking pairs of phrases, the number of bits required for the predicted symbol with the explicit coding, the number of bits required using predictive coding, and the total savings. The ellipsis between the two phrases in the first column represents the variable content between the keywords.... In PAGE 13: ... Note, however, that the performance of PPMC and WORD could no doubt be improved by utilizing models of higher order. Table2 ranks possible generalisations in order of their usefulness. Predictions 1, 2, 3 and 6 encourage generalisation of codes, which has been performed.... ..."

"... In PAGE 8: ... The most reliable class assignments correspond to a cluster containing mostly tissues of one tumor class, with a surrounding neighborhood of nodes also containing mostly tissues of the same tumor class. Table2 lists the class prediction assignments, grouped according to each of the 14 tumor classes. Each column Table 1 Tissue predictions based on original data Tissue Preference Hits Total Tm-BRa 0.... In PAGE 14: ... The highest accuracies were found for the leukemia training set, with 26 of the 38 data vectors having correct assignments. In general, however, the predic- tion accuracies were 50% poorer than those reported in Table2 . Establishing the reasons for these differences is difficult; however, the limited number of genes in each test dataset most likely contributes to the poorer classification accuracies.... ..."

"... In PAGE 21: ... For this reason we expanded regression to include seven transaction year indicator variables, TransYear(J,t) for J=1995 to 2001. Table2 reports the results of this regression. Although the transactions year indicator variables for 2000 and 2001 are estimated to be very large and positive, the estimates of (00 , (01 and 800 and 801 are all positive, different from Table 1, only 800 and 801 are statistically significantly different from zero.... ..."