Statistical Analysis of a Multi-Criteria Assessment of Intelligent Traffic Systems for the Improvement of Road Safety

Tables index

Veiw figure View Table

3. Analysis

As a first step to the road safety evaluation of ITS system groups, experts assessed the described categoriesaccording to the determined criteria and scale ^[16]. When performing the analysis we worked with the average scores summarized in Table 2.. Results of the evaluation provided basis for the fundamental Table of the KIPA analysis and preparation of the cluster analysis.

Table 2. Average of the expert assessments to the ITS system groups

Download as

Tables index

Veiw figure View Table

In order to study the appropriateness of analyses based on average expert opinion values, we examined relative dispersion values of the average scores and weights related to the ITS system groups and the study aspects, these are summarized in Table 3 and Table 4.

Table 3. Relative dispersions of weight numbers given to the assessment criteria

Download as

Tables index

Veiw figure View Table

Table 4. Relative dispersion of the average scores given to the ITS system groups

Download as

Tables index

Veiw figure View Table

In some cases, relative dispersions had values higher than the expected number of 0.2, demonstrating the heterogeneous structure of the scoring system, however, no extreme values were observed, so we accepted the dispersions in our primary studies and considered the sample well characterized by the average values.

The KIPA procedure is suitable for comparison of complex systems, by using it we will be able to rank the determined 10 ITS system groups supporting road safety, by considering all the studied aspects, thereby we will be able to identify the types of provisions application of which the most advantageous effects can be reached ^[8]. Characterization of the alternatives takes place on scales formed based on the assessment criteria, based on the principle of even comparison ^[4]. Basis for the qualification of the alternatives according to the scale formed is the average of the expert scores. Main steps of the procedure are as follows:

1. designing the measurement scales of the assessment factors (by considering the weights);

2. preparation of the basic table of the KIPA method (by projecting the expert qualifications of the alternatives to the scale described previously);

3. preparation of the KIPA matrix (comparison in pairs);

4. specifying limit values of preference and disqualificance;

5. drawing an assorting graph;

6. determination of the preference order.

The scales of measurement can be formed based on weight factors of the assessment criteria. When preparing the scales, a value of 50 was assigned to the score 5 in each case. Increments of the scale corresponding to the aspect with the lowest weight factor are size 1, and increments for scales of the other criteria increased according to the weigh factor differences. Values of the measurement scales assigned to the integer scores are shown in Table 5.

Table 5. Measurement scales for the assessment factors of the KIPA analysis

Download as

Tables index

Veiw figure View Table

Basic table of KIPA can be prepared based on the expert score averages described in Table 2, by using the measurement scale. Values assigned to the non-integer scores have been calculated by means of interpolation, by using interpolation based on values assigned to the integer scores. Values of the basic table were summarized in Table 6.

Table 6. Basic Table of the KIPA analysis

Download as

Tables index

Veiw figure View Table

In order to generate the KIPA matrix the first steps were calculation of the c_ij preferency (preference) and d_ij disqualificancy (disadvantage), based on comparisons in pairs.

The preference index provides information on the advantage of the ith alternative compared to the jth one; this index was calculated for each respect. Its value arises by summarizing those assessment criteria (in per cent) for which the particular ITS system group is preferred or neutral (its assigned value in the basic table is greater than or equals the value of the counterpart) over against the compared alternative.

The disadvantage indices were also calculated for each ij relations, however, to their determination only that assessment aspect was taken into account, for which the preference intensity is the highest. Thus, in the fist step of generating value of the disadvantage indices the assessment aspect is selected which meets the following two criteria:

• according to the particular aspect, value assigned to the jth alternative is higher the the one assigned to the ith element.

• absolute value of the difference between the values assigned to the ith and jth alternatives.

Absolute value of the difference of the numbers according to the aspects selected this way is the maximum scale difference wherein the studied alternative i is disadvantageous compared to alternative j. To calculate the disqualificancy indices, his value is divided by the volume of the maximum scale (in the present case this is the scale according to the road safety impact), then the result is multiplied by 100 to obtain the per cent form.

The KIPA matrix can be generated by the use of the preferency and disqualificancy indices (Figure 1). First row and column of the matrix contains numbers of the alternatives (i.e. numbers of the studied ITS system groups); in the top of the cell formed by the intersection of row i and column j (in bold italics) the c_ij advantage index, and in the bottom part (in italics) the d_ij disadvantage index is shown.

Download as

Veiw figureFigures index

NEW
View larger figure in new window

View next figure

Figure 1. The KIPA matrix

Determination of the order of preference and ranking of the ITS system groups were performed based on the matrix data. The preferency threshold value was taken at a 60% level (c_i_j≥60%) and the disqualificancy threshold value at 30% level (d_i_j≤ 30%) into account, these limits were satisfied by altogether 34 pairwise comparisons. This made possible to draw the assortational graph (Figure 2); nodes of the graph were formed by the ITS groups studied, edges between the compared pairs of elements are directed, they start from the preferred alternative and point to the other element. The graph can be applied to the determination of the order f versions and the demonstration of the results (the alternative from which all arrows point “outwards” is the best, while the version towards which the highest number of arrows point is the worse).

Download as

Veiw figureFigures index

View next figure

Figure 2. Assortational graph

Rank order of the alternatives became readable by the use of the assortational graph, order of the studied ITS system groups are summarized in Table 7. The order shows solutions for which their introduction, spreading and supporting is recommended the best.

Introduction of the in-vehicle active road safety systems (group 9) proved to be the most advantageous type of provision, edges setting out from which point to all the other groups. These ITS solutions significantly improve the road safety, they have relatively low expenses and present only few difficulties, and at the same time more and more traffic participants can be reached by them as technology develops. Also extremely good results were reached by the group of systems assisting the compliance to the Highway Code rules (Group 2), which is clearly preferred to all of the groups except Group 9 and 6. In addition to this, first part of the rank order contained the systems for prediction of the dangerous situations (Group 4), the systems assisting the rescue activities (Group 5) and the intervention systems depending on the drivers’ condition (Group 1). The worst result was shown by the real-time, personalized continuous following of the road behaviour, supplying recommendation and assistance (Group 3); the reason for this is that only a moderate improvement can be expected in its effect on the road safety which was taken with the highest weight into account, moreover, its implementation could involve significant social/legal difficulties, as the related information comprises highly sensitive personal data. Group 7, i.e. the traffic control systems are also not preferred to any other groups of system, primary function of them is not providing for road safety or protection, and their implementation costs showed disadvantageous scores, too. In addition to these, second part of the order of preference consists of the systems generating the road safety characteristics of the infrastructure (Group 8, preferred to Groups 3, 10 and 7), the unified system of EU drivers’ licenses and registration (Group 6, preferred only to Group 3), and the monitoring systems to be compulsorily used inside the vehicle (Group 10, also preferred only to Group 3).

Table 7. Order of preference of the studied ITS system groups

Download as

Tables index

Veiw figure View Table

In relation to the 10 ITS system groups formed based on the ITS solutions improving the road safety we performed a cluster analysis. Clustering is a group-forming procedure which – with taking several aspects into account – the similar elements are classified into the same group and the elements differing from each other to other, differing groups. Aim of the analysis was in part supporting results of the KIPA procedure expressly providing an order of preference, and in part identification of groups suitable for granting and consisting of advantageous measures; this can help decision-makers in the formation of provision packages ^[17].

Due to the high number of aspects to be taken into account, formation of groups is not an unambiguous task and various methods, based on different distance or similarity measures, can be applied to perform the procedure. An ideal clustering algorithm does not exist, as there is no objective measure to the comparison of results ^[2], which algorithm is practical to choose depends on the character of each application ^[13].

In our study, to the cluster analysis we applied the Ward method listed among the hierarchical procedures, regarding that this procedure is very widespread in the economical applications, it can be well interpreted from practical aspects, too, and it results in groups roughly the same size ^[11]. The method is based on a merger, i.e. initially it regards each element a separate cluster and connects them to form increasingly larger groups. For the Ward procedure, the average (or the sum of the squared variances from the average) is calculated for the points inside the cluster and that point or cluster is included in the formation of the larger cluster, for which the inclusion would bring about the higher increase in the sum of squared variances. This method is thus applicable to our study well, as the ITS system groups forming the database are characterized by the average of points given in the expert-performed classification (1-10 scale), thus these numerical distances can be applied to measure similarity of the elements, the initial variables are not correlated and there are no exceling data.

Cluster analysis was performed by means of the IBM SPSS Statistics program. Upon the procedure, the program classified the specified 10 elements into 2 to 4 clusters, of which groups with the most homogeneous number of elements were formed for the version with 3 clusters. Individual clusters comprised the following ITS system groups:

1. Cluster: ITS system groups Nos 1, 7, 8 and 10

2. Cluster: ITS system groups Nos 2, 4, 5 and 9

3. Cluster: ITS system groups Nos 3 and 6

Values to the index numbers of the generated clusters were summarized in Table 8.

Table 8. Index numbers of the ward procedure generating the 3 clusters

Download as

Tables index

Veiw figure View Table

Each cluster can be assessed based on the averages according to the various variables and by using their differences from the averages relative to all the elements. The lowest differences were seen at the scores related to the implementation expenses; more significant differences were found among the other 4 aspects, therefore they also had a higher role in the group formation. Steps of the clustering method are demonstrated in Figure 3.

Download as

Veiw figureFigures index

Figure 3. Dendogram of the cluster analysis [1. Intervention Systems Influenced by the Condition of the Driver, 2. Systems Assisting to Keep the Highway Code, 3. Individualized real-time continuous following and assisting of the participant’s traffic behaviour, 4. Forecast of dangerous traffic situations, 5. Systems assisting the rescue activities, 6. Unified electronic driving licenses and registration system in the EU, 7. Traffic control systems, 8. Generating road safety characteristics of the road infrastructure, 9. Active in-vehicle road safety systems, 10. In-vehicle monitoring systems assisting the road safety and to be used compulsorily]

The cluster-average-based assessment of the groups formed is summarized in Table9 (the + and - signs symbolize the direction and magnitude of deviations from the average of all elements).

Table 9. Characterization of the clusters in comparison to the average of all elements

Download as

Tables index

Veiw figure View Table