Page 140 - 360.revista de Alta Velocidad - Nº 5

Grande, Zacarías. Blanco López, Marta. García Tamames, Alberto. Castillo, Enrique.

2.4 Backward analysis

The Bayesian network also allows to determine the causes of a given incident by proceeding
backwards using available information and modifying the probabilities accordingly. Figure 8
shows the conditional probability tables before any information is supplied. For example, the
probability of a severe accident number 33 is as low as 9.01 x 10 , signal 31 is in red 15% of the
-6
times, the probability of the driver to make an erroneous decision 14 is r, not obeying the sign,
-5
the speed was 220 km/h, there was a driver’s erroneous decision14 “error I” is 6.01 x 10 , and
-7
the probability of the driver to be distracted was 2.84 x 10 .

Figure 8 Conditional probability tables before any information is supplied.

However, after an accident 33 has occurred and evidence is obtained about the no occurrence
of technical failures 16 and 30, their probabilities become one, as indicated in Figure 9 The
Bayesian network techniques permit recalculating the probabilities of all other variables and
consequently, provide the probabilities of any other event having information about the causes
of accident 33. For example, Figure 9 informs us that signal 31 was in red, the driver made an
error, not obeying the sign, the speed was 220 km/h, there was a driver’s erroneous decision14
“error I”, and this could be due to a distracted state or to an attentive state, but the later
seems to be more probable.

138 360.revista de alta velocidad