Enginius

Segmentation
Shivaan Chopra

Copyright (c) 2020, DecisionPro Inc.

The segmentation data has been scaled column wise
4-segment solution
The ideal number of segments is a function of statistical fit (what the data say), managerial relevance (what makes th
most sense from a managerial point of view), and targetability (can the segments be easily targeted).
When the three criteria do not perfectly converge, selecting the right number of segments becomes a judgment call.
You have decided to perform the analysis with 4 segments.
The segmentation method relies on the hierarchical clustering approach. This approach generates a dendrogram that
display next.

Dendrogram
The dendrogram represents the grouping process of observations into clusters. The chart reads from bottom (all initia
observations are separated) to top (all observations are clustered into one unique segment).
The height represents the distance between the two groups of observations being merged at each step. If two very d
groups are being merged, this will create a 'jump' in the dendrogram, indicating that it might be wise to stop the
clustering process before.
Dendrogram. The dendrogram is a tree diagram to illustrate the arrangement of clusters produced by hierarchical clustering, and

Scree plot
The screeplot displays, for each cluster solution, a measure of within-cluster heterogeneity. If clusters group
observations that are widely different (which will happen if the number of clusters is too small to capture the
variability in the data), the value will be high.
A good cluster solution might be where the screeplot displays an 'elbow', that is, where increasing the number of
clusters beyond a certain point does not dramatically decreases within-cluster heterogeneity.
The measure displayed in the screeplot is related, but not equivalent, to the distance reported in the dendrogram.
Scree plot. The scree plot compares the sum of squared error (SSE) for each cluster solution. A good cluster solution might be wh
From a statistical point of view, the SSE reported in the screeplot is computed as the sum of squared error between e
observation and its cluster centroid (or center), summed over all the observations.
 y), managerial relevance (what makes the
ments be easily targeted).
r of segments becomes a judgment call.

s approach generates a dendrogram that we

rs. The chart reads from bottom (all initial

nique segment).
being merged at each step. If two very distant
ting that it might be wise to stop the
sters produced by hierarchical clustering, and how the observations are incrementally clustered together.

heterogeneity. If clusters group

usters is too small to capture the

at is, where increasing the number of

er heterogeneity.
distance reported in the dendrogram.
solution. A good cluster solution might be when the SSE slows dramatically, creating an 'elbow'. Such elbow does not always exist.
d as the sum of squared error between each
Segment size

Segment size
Population Segment 1
Size 317 82
Relative size 100% 26%

Segment description

Segment description
Average value of each segmentation variable, overall for each segment (centroid). Segmentation variables that are statistically di
Population Segment 1
Rich full-bodied 4.77 3.48
Light beer 3.72 2.23
No aftertaste 4.56 4.26
Refreshing 5.02 5.65
Goes down easily 5.17 5.21
Gives a buzz 3.39 2.82
Good taste 3.000 1.268
Low price 3.91 3.41
Good value 4.65 4.44
From country with brewing tradition 3.82 3.04
Attractive bottle 3.00 2.44
Prestigious brand 3.20 1.87
High quality 4.48 3.06
Drink at picnics 4.56 4.62
Masculine 2.67 1.54
For young people 2.49 1.33
Drink with friends 4.70 4.57
Drink at home 4.34 4.20
To serve dinner guests 4.92 5.06
For dining out 5.03 4.91
Drink at bar 4.35 3.57
Segment differences per segment. Cell colors indicate to what extent a segment is statistically different from the rest of the popu

Segment space
The chart below is a graphical representation of the various segments, segment members, and segmentation variable
is obtained by outputing the first two dimensions of a principal component analysis performed on the (standardized)
segmentation data, on top of which segment information has been overlayed.
Because only the first two dimensions of the PCA are displayed, and these two dimensions capture only 34.1% of the
variance in the data, some differences between segments might not appear here. Note that segmentation variables w
variance, if any, have been excluded.
Two clusters who appear to overlap in the first two dimensions might actually be distinct on other dimensions.
Consequently, this chart is a useful guide, especially to see which segmentation variables are correlated, but may be
misleading if used to select the optimal number of segments.
Segment space. Spatial representation of segments and segmentation variables, using principal component analysis.

Segment membership

Segment membership
Segment to which each member of the population belongs to. The complete membership list is only available in the Excel format
Segment
6861 1
4129 1
4393 2
445 3
7393 4
964 3
6773 4
461 2
7156 4
5785 4
4946 3
428 2
4465 1
2478 1
6321 2
4263 1
4305 2
5074 4
3004 4
702 3
6778 4
5542 2
7247 4
6024 2
6172 2
2590 1
6293 3
4485 3
5520 1
5837 4
1868 2
5996 1
2588 1
387 2
5660 3
6758 1
7187 3
993 2
301 1
7018 1
3570 1
4829 2
5575 1
6100 1
3950 2
1241 2
1729 2
4377 4
4448 2
7049 3
4970 2
7201 4
5555 2
6021 3
3528 3
1129 4
6502 2
2911 1
2899 1
6421 1
616 4
141 1
464 2
1173 1
525 4
5646 3
166 2
382 1
4953 2
5789 3
3655 1
1574 2
3642 2
6450 2
3774 3
3711 2
5691 4
3602 4
2316 4
1694 4
807 1
7326 1
3485 4
3125 1
1035 2
763 3
3014 3
707 3
5655 2
5320 4
4413 4
4537 2
534 1
1275 1
388 3
7211 2
5573 2
4668 1
3490 2
4350 3
4614 2
3819 1
6855 2
2024 2
3373 3
5701 3
4174 1
2810 1
3996 1
7447 2
4491 4
3595 4
308 1
3514 3
71 1
3953 3
6048 4
6494 2
1223 4
3967 2
4643 1
5824 4
1322 4
3520 3
6157 3
3028 2
5870 4
3800 1
6068 4
5637 2
7460 1
7173 1
6910 1
1555 4
1743 2
3833 3
5043 4
6814 3
6605 4
7152 4
5850 2
863 1
7487 4
2298 2
4181 4
748 2
3507 1
5262 2
3962 1
1242 2
2338 1
389 3
62 3
5506 1
4388 4
4480 3
5543 2
2062 2
4841 4
5839 3
7308 3
4956 2
6551 1
6312 2
1696 2
5856 1
2403 2
7413 1
5400 1
1176 2
1595 3
717 2
574 4
2523 4
5035 1
3154 3
2029 1
4262 3
7134 2
489 3
6674 2
6550 3
7415 2
2009 4
6423 4
4195 4
1232 2
6831 3
2054 3
985 3
708 4
4670 3
6351 2
2013 1
6316 2
7124 1
2569 3
1676 2
6237 4
2936 4
3666 4
6669 2
2857 1
5855 2
2474 4
3617 3
1324 1
6221 2
1283 3
3212 4
7355 2
3225 1
6817 1
6858 1
6970 2
4420 3
3762 4
4545 2
2459 2
3842 1
7165 2
2687 1
2819 1
5429 3
7501 2
1733 3
6641 2
4323 4
4653 2
4091 1
206 1
4459 2
2708 2
6189 1
3535 1
599 3
385 1
2128 4
6225 2
903 1
4743 2
7191 4
6352 1
6881 1
4967 3
4762 4
1730 3
5913 2
298 3
7192 3
3483 3
5862 1
2997 1
5698 1
2052 4
5481 4
3733 1
5353 1
2787 2
662 3
4136 3
7304 2
4626 3
5450 2
5628 3
4295 3
6612 2
6013 2
3614 2
5045 3
4794 3
4424 2
6373 4
1712 1
4421 2
6575 1
4498 1
5406 1
7519 2
2204 3
7169 3
73 3
5719 4
5832 3
7172 4
5803 2
4382 2
7388 1
5794 2
3011 2
7189 2
2587 2
4908 3
879 1
6291 2
6148 4
2932 3
5786 2
7128 4
997 2
984 1
6702 1
3974 2
3970 2
5633 3
7180 4
3214 2
7469 4
381 3
4196 4
5619 3
7500 3
6071 2
2319 2
6145 1
3460 4
5456 4
 Segment 2 Segment 3 Segment 4
100 70 65
32% 22% 0.20504731861199

egmentation variables that are statistically different from the rest of the population are highlighted in red (lower) or green (higher).
Segment 2 Segment 3 Segment 4
6.08 2.30 7.0307692307692
4.27 3.67 4.8153846153846
4.46 3.46 6.2923076923077
5.23 2.00 7.1692307692308
5.46 2.91 7.0923076923077
3.44 2.63 4.8769230769231
3.280 0.443 7.5076923076923
4.12 3.13 5.0615384615385
4.42 3.26 6.7692307692308
4.42 2.54 5.2461538461538
2.67 2.77 4.4461538461538
3.81 2.27 4.9384615384615
6.02 1.53 7.0615384615385
5.21 2.49 5.7384615384615
2.67 2.13 4.6769230769231
2.58 2.29 4.0307692307692
6.26 1.36 6.0461538461539
5.02 1.60 6.4153846153846
6.17 1.46 6.5384615384615
6.17 1.67 7.0461538461539
5.69 1.56 6.2615384615385
 atistically different from the rest of the population on each segmentation variable.

ent members, and segmentation variables. It

nalysis performed on the (standardized)

wo dimensions capture only 34.1% of the

here. Note that segmentation variables with no

y be distinct on other dimensions.

on variables are correlated, but may be
ng principal component analysis.

ership list is only available in the Excel format.

Descriptors
This table reports the discriminant variable (or descriptor) averages of each segment. The more differences can be
found, the easier it will be to predict segment membership based on discriminant data alone.

Discriminant data per segment

Average value of each discriminant variable, overall and within each cluster. Discriminant variables that are statistically different
Population Segment 1
Weekly consumption 9.42 11.54
Age (1-7) 4.77 4.51
Income (1-7) 5.45 5.45
Education (1-6) 4.47 4.66
Sex (male=1) 1.12 1.10
Adapt to new situations 3.50 3.61
Make friends easily 3.31 3.29
Do not like to be tied to timetable 3.56 3.50
Like to take chances 3.13 3.28
Like to travel abroad 3.31 3.34
Like ethnic food 3.47 3.56
Knowledgeable about beer 2.95 3.04
Discriminant differences per segment. Cell colors indicate to what extent the distribution of a discriminant variable in a segment

Discriminant space
The chart below is a graphical representation of the various segments, segment members, and discriminant variables
is obtained by outputing the first two dimensions of a principal component analysis performed on the (standardized)
discriminant data, on top of which segment information has been overlayed.
Because only the first two dimensions of the PCA are displayed, and these two dimensions capture only 30.3% of the
variance in the data, some differences between segments might not appear here. Note that discriminant variables wi
variance, if any, have been excluded.
If two or more segments fully overlap, it is unlikely that they could be clearly separated based on discriminant data
alone.
However, two segments that seem to overlap on two dimensions may be more clearly separated on other dimension
Consequently, the confusion matrix is a better guide to assess the quality of segment discrimination.
Discriminant space. Spatial representation of segments and discriminant variables, using principal component analysis.
segment. The more differences can be
inant data alone.

inant variables that are statistically different from the rest of the population are highlighted in red (lower) or green (higher).
Segment 2 Segment 3 Segment 4
8.74 9.19 8.0461538461538
4.89 4.94 4.7230769230769
5.58 5.51 5.1846153846154
4.52 4.21 4.4461538461538
1.13 1.17 1.0769230769231
3.47 3.56 3.3538461538461
3.35 3.39 3.2
3.63 3.54 3.5384615384615
3.03 3.19 3.0461538461538
3.32 3.23 3.3230769230769
3.49 3.49 3.3076923076923
2.96 2.86 2.9230769230769
ution of a discriminant variable in a segment is statistically different from the rest of the population.

ent members, and discriminant variables. It

nalysis performed on the (standardized)

wo dimensions capture only 30.3% of the

here. Note that discriminant variables with no

separated based on discriminant data

ore clearly separated on other dimensions.

segment discrimination.
using principal component analysis.
Introduction
Often, segmentation variables may not be available to managers, but descriptors may be.
In this section, we explore whether descriptors alone could predict segment membership with sufficient accuracy. Th
confusion matrix and hit rates (reported below) indicate whether the model is accurate enough.
For discriminant analysis, Enginius uses a multinomial logit model (similar to the one used to predict 'choices between
multiple alternatives (A/B/C)' in the predictive modeling module.
The largest segment is selected as the default option (dummy), and the model identifies which discriminant variables
the most significant to predict cluster memberships. If a discriminant variable is highly predictive, its p-values will
be close to zero, and the cells will appear in green (or red).

Model coefficients

Model parameters
Segment 2 is the model baseline.
Segment 1 Segment 3
(Intercept) 0.0738 -0.4419
Weekly consumption 0.01870 0.00385
Age (1-7) -0.1543 0.0558
Income (1-7) -0.02539 0.00585
Education (1-6) 0.0927 -0.1606
Sex (male=1) -0.211 0.383
Adapt to new situations 0.373 0.297
Make friends easily -0.2436 -0.0303
Do not like to be tied to timetable -0.542 -0.402
Like to take chances 0.542 0.445
Like to travel abroad -0.0951 -0.1777
Like ethnic food 0.1342 0.0258
Knowledgeable about beer -0.0770 -0.1325

P-values

p-values
Probabilities that parameter estimates are different from zero only by chance.
Segment 1 Segment 3
(Intercept) 0.967 0.809
Weekly consumption 0.253 0.835
Age (1-7) 0.198 0.652
Income (1-7) 0.816 0.960
Education (1-6) 0.450 0.173
Sex (male=1) 0.675 0.409
Adapt to new situations 0.192 0.305
Make friends easily 0.258 0.892
Do not like to be tied to timetable 0.044 0.151
Like to take chances 0.024 0.061
Like to travel abroad 0.618 0.354
Like ethnic food 0.576 0.914
Knowledgeable about beer 0.686 0.486

Confusion matrix
The confusion matrix compares actual segment membership (obtained from the segmentation analysis and the origin
segmentation variables) and predicted segment membership (obtained from the discriminant analysis and the descrip
alone). When actual and predicted segment memberships coincide, the diagonal elements will be comparatively large
indicating that the discriminant model is accurate.

Confusion matrix (count)

The model has correctly classified 115 of the 317 observations. The off-diagonal elements are classification errors.
Predicted 1 Predicted 2
Segment 1 30 34
Segment 2 25 55
Segment 3 15 35
Segment 4 11 33
Total 81 157

Confusion matrix (%)

The global hit rate of the model is 36%. The diagonal elements represent segment-specific hit rates.
Predicted 1 Predicted 2
Segment 1 37% 41%
Segment 2 25% 55%
Segment 3 21% 50%
Segment 4 17% 51%
Confusion matrix (plot). Graphic representation of the confusion matrix: actual segment membership versus predicted segment m

Model predictions

Model predictions (in-sample)

This table details the probabilities of each member of the segmentation dataset to belong to each cluster (as predicted by the dis
Prob(cluster 1) Prob(cluster 2)
686100% 24% 35%
412900% 28% 21%
439300% 36% 31%
44500% 23% 42%
739300% 24% 26%
96400% 32% 23%
677300% 23% 31%
46100% 11% 32%
715600% 22% 29%
578500% 29% 25%
494600% 8% 34%
42800% 27% 31%
446500% 27% 18%
247800% 41% 19%
632100% 54% 18%
426300% 26% 26%
430500% 21% 40%
507400% 15% 41%
300400% 14% 31%
70200% 17% 38%
677800% 24% 33%
554200% 11% 27%
724700% 21% 28%
602400% 41% 27%
617200% 26% 26%
259000% 25% 23%
629300% 14% 35%
448500% 26% 36%
552000% 58% 17%
583700% 26% 30%
186800% 30% 26%
599600% 19% 26%
258800% 26% 40%
38700% 22% 41%
566000% 42% 24%
675800% 27% 32%
718700% 18% 44%
99300% 50% 18%
30100% 16% 51%
701800% 37% 29%
357000% 55% 24%
482900% 24% 40%
557500% 36% 27%
610000% 21% 27%
395000% 22% 41%
124100% 27% 37%
172900% 12% 43%
437700% 30% 27%
444800% 33% 30%
704900% 26% 35%
497000% 29% 34%
720100% 10% 35%
555500% 23% 30%
602100% 19% 35%
352800% 9% 53%
112900% 30% 33%
650200% 36% 30%
291100% 12% 34%
289900% 28% 33%
642100% 22% 36%
61600% 10% 46%
14100% 12% 41%
46400% 9% 32%
117300% 10% 40%
52500% 23% 33%
564600% 14% 31%
16600% 12% 42%
38200% 32% 27%
495300% 26% 17%
578900% 34% 17%
365500% 41% 21%
157400% 16% 27%
364200% 30% 27%
645000% 16% 30%
377400% 34% 22%
371100% 16% 39%
569100% 21% 36%
360200% 41% 22%
231600% 26% 31%
169400% 18% 45%
80700% 17% 47%
732600% 31% 25%
348500% 36% 23%
312500% 27% 28%
103500% 38% 28%
76300% 19% 13%
301400% 23% 28%
70700% 29% 27%
565500% 18% 40%
532000% 12% 43%
441300% 28% 29%
453700% 17% 22%
53400% 18% 33%
127500% 26% 40%
38800% 7% 53%
721100% 12% 25%
557300% 21% 38%
466800% 20% 38%
349000% 12% 35%
435000% 26% 32%
461400% 35% 33%
381900% 17% 37%
685500% 36% 25%
202400% 27% 34%
337300% 24% 31%
570100% 30% 32%
417400% 39% 28%
281000% 16% 20%
399600% 22% 33%
744700% 26% 30%
449100% 19% 30%
359500% 37% 28%
30800% 37% 23%
351400% 29% 23%
7100% 26% 37%
395300% 24% 36%
604800% 45% 21%
649400% 14% 40%
122300% 21% 42%
396700% 12% 47%
464300% 26% 24%
582400% 16% 32%
132200% 12% 31%
352000% 20% 40%
615700% 34% 27%
302800% 13% 48%
587000% 24% 39%
380000% 25% 37%
606800% 48% 10%
563700% 34% 25%
746000% 41% 21%
717300% 25% 33%
691000% 28% 29%
155500% 20% 29%
174300% 17% 38%
383300% 25% 35%
504300% 32% 26%
681400% 24% 19%
660500% 24% 22%
715200% 24% 38%
585000% 26% 30%
86300% 37% 23%
748700% 24% 30%
229800% 37% 33%
418100% 18% 36%
74800% 9% 29%
350700% 47% 25%
526200% 26% 26%
396200% 37% 34%
124200% 15% 37%
233800% 39% 27%
38900% 21% 38%
6200% 14% 29%
550600% 49% 17%
438800% 19% 35%
448000% 44% 24%
554300% 24% 34%
206200% 31% 31%
484100% 15% 38%
583900% 11% 36%
730800% 10% 47%
495600% 30% 35%
655100% 46% 18%
631200% 20% 40%
169600% 38% 27%
585600% 36% 16%
240300% 29% 31%
741300% 17% 36%
540000% 30% 32%
117600% 24% 32%
159500% 18% 38%
71700% 6% 48%
57400% 15% 31%
252300% 25% 25%
503500% 22% 34%
315400% 29% 30%
202900% 38% 24%
426200% 21% 33%
713400% 10% 43%
48900% 60% 13%
667400% 12% 44%
655000% 25% 31%
741500% 57% 20%
200900% 23% 36%
642300% 26% 36%
419500% 21% 24%
123200% 33% 32%
683100% 17% 46%
205400% 62% 18%
98500% 32% 21%
70800% 21% 38%
467000% 17% 35%
635100% 15% 48%
201300% 25% 25%
631600% 16% 36%
712400% 32% 32%
256900% 26% 43%
167600% 22% 27%
623700% 26% 36%
293600% 39% 25%
366600% 13% 37%
666900% 21% 27%
285700% 34% 30%
585500% 12% 37%
247400% 21% 32%
361700% 27% 21%
132400% 28% 32%
622100% 25% 35%
128300% 16% 22%
321200% 22% 29%
735500% 41% 36%
322500% 25% 36%
681700% 44% 21%
685800% 54% 13%
697000% 27% 27%
442000% 55% 13%
376200% 18% 35%
454500% 13% 32%
245900% 12% 36%
384200% 36% 31%
716500% 16% 41%
268700% 16% 23%
281900% 29% 33%
542900% 23% 30%
750100% 28% 29%
173300% 34% 32%
664100% 22% 38%
432300% 24% 32%
465300% 30% 42%
409100% 34% 33%
20600% 48% 27%
445900% 45% 23%
270800% 13% 51%
618900% 58% 10%
353500% 30% 37%
59900% 18% 35%
38500% 22% 47%
212800% 41% 27%
622500% 44% 17%
90300% 31% 32%
474300% 8% 44%
719100% 31% 18%
635200% 28% 32%
688100% 40% 22%
496700% 29% 35%
476200% 31% 31%
173000% 24% 31%
591300% 32% 31%
29800% 24% 29%
719200% 22% 30%
348300% 15% 24%
586200% 22% 32%
299700% 36% 27%
569800% 16% 37%
205200% 21% 36%
548100% 16% 36%
373300% 12% 32%
535300% 60% 18%
278700% 11% 36%
66200% 27% 40%
413600% 32% 30%
730400% 44% 24%
462600% 25% 28%
545000% 20% 42%
562800% 30% 24%
429500% 18% 34%
661200% 25% 41%
601300% 21% 42%
361400% 18% 45%
504500% 19% 35%
479400% 39% 26%
442400% 16% 27%
637300% 9% 58%
171200% 16% 48%
442100% 15% 37%
657500% 31% 21%
449800% 49% 26%
540600% 26% 20%
751900% 23% 34%
220400% 10% 45%
716900% 32% 25%
7300% 20% 29%
571900% 23% 31%
583200% 16% 22%
717200% 14% 49%
580300% 29% 36%
438200% 33% 20%
738800% 21% 41%
579400% 27% 28%
301100% 36% 26%
718900% 33% 29%
258700% 27% 33%
490800% 48% 18%
87900% 25% 39%
629100% 25% 35%
614800% 33% 24%
293200% 12% 40%
578600% 40% 24%
712800% 22% 26%
99700% 12% 45%
98400% 28% 27%
670200% 47% 19%
397400% 12% 27%
397000% 40% 20%
563300% 33% 29%
718000% 22% 30%
321400% 12% 41%
746900% 19% 30%
38100% 31% 28%
419600% 15% 44%
561900% 25% 38%
750000% 21% 39%
607100% 21% 41%
231900% 13% 51%
614500% 28% 34%
346000% 34% 32%
545600% 18% 27%
ptors may be.
membership with sufficient accuracy. The
is accurate enough.
the one used to predict 'choices between

el identifies which discriminant variables are

e is highly predictive, its p-values will

Segment 4
3.6752448117395
-0.014206784116895
-0.050670139697248
-0.12519701129326
0.019778456087426
-0.66900518842226
-0.23558312882646
-0.15358065224408
-0.24436589265983
0.17873469652749
0.1036731651552
-0.29474006075355
-0.040855236012989

Segment 4
0.043603601936111
0.49427840861891
0.68835693555349
0.27031514075597
0.87366391797036
0.24274722003337
0.40133100149874
0.49303052454743
0.39773334350524
 0.46780934868293
0.61979890231218
0.20721263487867
0.83971214143618

the segmentation analysis and the original

the discriminant analysis and the descriptors
onal elements will be comparatively large,

ments are classification errors.

Predicted 3 Predicted 4 Total
7 11 82
11 9 100
15 5 70
6 15 65
39 40 317

pecific hit rates.

Predicted 3 Predicted 4 Total
9% 13% 1
11% 9% 1
21% 7% 1
9% 23% 1
ment membership versus predicted segment membership. Bubbles in the diagonale represent correct classification.

elong to each cluster (as predicted by the discriminant model and the descriptors alone). The segment with the highest probability is retained, and is co
Prob(cluster 3) Prob(cluster 4) Predicted Actual Correct
17% 0 2 1 0
42% 0 3 1 0
17% 0 1 2 0
21% 0 2 3 0
19% 0 4 4 1
37% 0 3 3 1
13% 0 4 4 1
27% 0 2 2 1
21% 0 2 4 0
14% 0 4 4 1
25% 0 2 3 0
30% 0 2 2 1
24% 0 4 1 0
19% 0 1 1 1
14% 0 1 2 0
15% 0 4 1 0
17% 0 2 2 1
16% 0 2 4 0
14% 0 4 4 1
31% 0 2 3 0
19% 0 2 4 0
18% 0 4 2 0
21% 0 4 4 1
16% 0 1 2 0
35% 0 3 2 0
22% 0 4 1 0
12% 0 4 3 0
20% 0 2 3 0
15% 0 1 1 1
19% 0 2 4 0
23% 0 1 2 0
45% 0 3 1 0
16% 0 2 1 0
19% 0 2 2 1
25% 0 1 3 0
20% 0 2 1 0
22% 0 2 3 0
17% 0 1 2 0
13% 0 2 1 0
22% 0 1 1 1
11% 0 1 1 1
16% 0 2 2 1
25% 0 1 1 1
14% 0 4 1 0
21% 0 2 2 1
25% 0 2 2 1
9% 0 2 2 1
13% 0 4 4 1
21% 0 1 2 0
22% 0 2 3 0
17% 0 2 2 1
48% 0 3 4 0
39% 0 3 2 0
33% 0 2 3 0
27% 0 2 3 0
14% 0 2 4 0
16% 0 1 2 0
17% 0 4 1 0
22% 0 2 1 0
22% 0 2 1 0
25% 0 2 4 0
33% 0 2 1 0
49% 0 3 2 0
30% 0 2 1 0
25% 0 2 4 0
28% 0 2 3 0
16% 0 2 2 1
15% 0 1 1 1
16% 0 4 2 0
40% 0 3 3 1
16% 0 1 1 1
 7% 1 4 2 0
18% 0 1 2 0
32% 0 3 2 0
26% 0 1 3 0
17% 0 2 2 1
30% 0 2 4 0
14% 0 1 4 0
17% 0 2 4 0
16% 0 2 4 0
18% 0 2 1 0
35% 0 3 1 0
17% 0 1 4 0
28% 0 3 1 0
20% 0 1 2 0
7% 1 4 3 0
35% 0 3 3 1
19% 0 1 3 0
8% 0 2 2 1
17% 0 2 4 0
15% 0 2 4 0
26% 0 4 2 0
16% 0 4 1 0
18% 0 2 1 0
29% 0 2 3 0
54% 0 3 2 0
11% 0 2 2 1
32% 0 2 1 0
46% 0 3 2 0
19% 0 2 3 0
19% 0 1 2 0
35% 0 2 1 0
21% 0 1 2 0
17% 0 2 2 1
35% 0 3 3 1
26% 0 2 3 0
20% 0 1 1 1
25% 0 4 1 0
30% 0 2 1 0
12% 0 4 2 0
28% 0 2 4 0
18% 0 1 4 0
17% 0 1 1 1
34% 0 3 3 1
22% 0 2 1 0
15% 0 2 3 0
21% 0 1 4 0
26% 0 2 2 1
15% 0 2 4 0
13% 0 2 2 1
28% 0 3 1 0
34% 0 3 4 0
32% 0 3 4 0
11% 0 2 3 0
29% 0 1 3 0
11% 0 2 2 1
17% 0 2 4 0
15% 0 2 1 0
32% 0 1 4 0
25% 0 1 2 0
21% 0 1 1 1
22% 0 2 1 0
28% 0 2 1 0
21% 0 4 4 1
17% 0 2 2 1
20% 0 2 3 0
27% 0 1 4 0
27% 0 4 3 0
23% 0 4 4 1
20% 0 2 4 0
25% 0 2 2 1
28% 0 1 1 1
13% 0 4 4 1
18% 0 1 2 0
35% 0 2 4 0
25% 0 4 2 0
16% 0 1 1 1
36% 0 3 2 0
17% 0 1 1 1
18% 0 2 2 1
25% 0 1 1 1
19% 0 2 3 0
31% 0 3 3 1
21% 0 1 1 1
11% 0 2 4 0
18% 0 1 3 0
28% 0 2 2 1
23% 0 2 2 1
15% 0 2 4 0
41% 0 3 3 1
33% 0 2 3 0
17% 0 2 2 1
22% 0 1 1 1
31% 0 2 2 1
19% 0 1 2 0
42% 0 3 1 0
18% 0 2 2 1
20% 0 2 1 0
18% 0 2 1 0
36% 0 3 2 0
25% 0 2 3 0
10% 0 2 2 1
13% 0 4 4 1
31% 0 3 4 0
23% 0 2 1 0
22% 0 2 3 0
23% 0 1 1 1
22% 0 2 3 0
19% 0 2 2 1
22% 0 1 3 0
16% 0 2 2 1
18% 0 2 3 0
13% 0 1 2 0
13% 0 2 4 0
21% 0 2 4 0
38% 0 3 4 0
22% 0 1 2 0
12% 0 2 3 0
13% 0 1 3 0
33% 0 3 3 1
13% 0 2 4 0
29% 0 2 3 0
24% 0 2 2 1
38% 0 3 1 0
21% 0 2 2 1
20% 0 2 1 0
16% 0 2 3 0
25% 0 2 2 1
17% 0 2 4 0
17% 0 1 4 0
6% 0 4 4 1
15% 0 4 2 0
21% 0 1 1 1
27% 0 2 2 1
29% 0 2 4 0
37% 0 3 3 1
10% 0 2 1 0
16% 0 2 2 1
42% 0 3 3 1
15% 0 4 4 1
11% 0 1 2 0
15% 0 2 1 0
23% 0 1 1 1
27% 0 1 1 1
32% 0 3 2 0
18% 0 1 3 0
14% 0 2 4 0
18% 0 4 2 0
21% 0 2 2 1
14% 0 1 1 1
24% 0 2 2 1
21% 0 4 1 0
23% 0 2 1 0
20% 0 2 3 0
15% 0 2 2 1
24% 0 1 3 0
19% 0 2 2 1
22% 0 2 4 0
14% 0 2 2 1
17% 0 1 1 1
16% 0 1 1 1
13% 0 1 2 0
24% 0 2 2 1
15% 0 1 1 1
16% 0 2 1 0
24% 0 2 3 0
15% 0 2 1 0
16% 0 1 4 0
22% 0 1 2 0
16% 0 2 1 0
16% 0 2 2 1
22% 0 1 4 0
26% 0 2 1 0
15% 0 1 1 1
21% 0 2 3 0
18% 0 2 4 0
31% 0 3 3 1
26% 0 1 2 0
31% 0 3 3 1
28% 0 2 3 0
33% 0 3 3 1
29% 0 2 1 0
16% 0 1 1 1
36% 0 2 1 0
21% 0 2 4 0
14% 0 2 4 0
23% 0 4 1 0
17% 0 1 1 1
29% 0 2 2 1
16% 0 2 3 0
29% 0 1 3 0
19% 0 1 2 0
17% 0 4 3 0
15% 0 2 2 1
32% 0 3 3 1
17% 0 2 3 0
17% 0 2 2 1
17% 0 2 2 1
14% 0 2 2 1
23% 0 2 3 0
25% 0 1 3 0
41% 0 3 2 0
19% 0 2 4 0
21% 0 2 1 0
24% 0 2 2 1
13% 0 4 1 0
14% 0 1 1 1
9% 0 4 1 0
28% 0 2 2 1
12% 0 2 3 0
32% 0 1 3 0
17% 0 4 3 0
19% 0 2 4 0
38% 0 3 3 1
16% 0 2 4 0
18% 0 2 2 1
35% 0 3 2 0
15% 0 2 1 0
26% 0 2 2 1
23% 0 1 2 0
17% 0 1 2 0
32% 0 2 2 1
15% 0 1 3 0
19% 0 2 1 0
25% 0 2 2 1
28% 0 1 4 0
31% 0 2 3 0
27% 0 1 2 0
23% 0 4 4 1
29% 0 2 2 1
23% 0 1 1 1
22% 0 1 1 1
17% 0 4 2 0
26% 0 1 2 0
23% 0 1 3 0
17% 0 4 4 1
20% 0 2 2 1
19% 0 4 4 1
19% 0 1 3 0
24% 0 2 4 0
19% 0 2 3 0
32% 0 2 3 0
27% 0 2 2 1
16% 0 2 2 1
18% 0 2 1 0
23% 0 1 4 0
34% 0 3 4 0
obability is retained, and is compared to the actual segment membership to measure model accuracy and classification errors.