The practice of using Big Data models is widespread implementing the procedures of information and technological support of processes occurring in urban resource social and communication networks. For Big Data, a list of characteristics is presented in the format 10V including volume, velocity, variety, validity, value, veracity, visibility, virtual, variability and valence. At the same time, the urgent task is to select the category of Big Data analytical processing tools for smart city needs. Further development of the Big Data concept leads to an expansion of the list of properties and a variety of tools for their analytical processing. It is suggested to use the expert estimation and hierarchy analysis methods proposed by T. Saati to select an effective Big Data analytical processing method for needs in a smart city. The procedures for this method are described below. The analysis was carried out among the following alternatives as managed machine learning, unmanaged machine learning, data mining, statistical analysis, data visualization. The obtained results show the highest efficiency of the method of managed machine learning. Furthermore, it should be noted that this method might be implemented in appropriate procedures considering the possible need to extend the sets of characteristics and their parameters. The proposed method allows to evaluate the advantages and disadvantages of the tools available on the market to work with big data in a situation when the city authorities decide on the need for appropriate investment.
The current tendency of reorganization of cities, their transformation into comfortable ones for residents and guests, implies the formation of a complex system of modern technologies and tools that facilitate interaction with state authorities, obtaining quality services in the field of health, energy, gas and water supply, as well as transport. One of the smart city projects' implementation areas is the implementation of IoT technologies and the formation of information platforms to process the accumulated data. 1.1
Problem Statement
The use of IoT technologies in smart city projects is being explored by many
scientists. It is noted in the paper [
The authors [
It is suggested [
However, the method of hierarchy analysis has found its use in projects of such type.
Some authors analyze why big data technologies with elements of machine
learning are effective to use in smart city projects [
Analyzing a wide variety of urban data sets, it can be stated that about 70 percent of
all data on resource social and communication networks [
It is suggested to formulate answers to this kind of questions using the hierarchy analysis method. A description of the procedures for using this method in the context of solving the problem of identifying the full set of Big Data characteristics and parameters in information systems of information and technological support of projects implementing in urban resource social and communication networks in smart cities is given below.
Simultaneously, it should be noted that this method has to be implemented in procedures considering the need to extend the sets of characteristics and their parameters. Experts on the use of Big Data models from European Council for Nuclear Research (CERN, Geneva) suggest that they will grow substantially to 20, 30 and even 100v in the near future.
The choice of Big data analytical processing methods for the needs in a smart city
[
Step 1. A hierarchy with a three-tier structure is built. The top level of the hierarchy is the goal to be decided (see Fig. 1).
Goal Criterion Criterion Criterion Criterion Criterion Criterion Criterion Criterion Criterion Criterion 1 2 3 4 5 6 7 8 9 10 Alternative MABD1
Alternative MABD2
Alternative MABD3
Alternative MABD4
Alternative MABD5 Fig. 1. The general scheme of the hierarchy analysis method of choosing a category for Big
Data analytical processing in a smart city Step 2. It is formed a criterion set at the second level, according to which alternative methods of analytical processing (MABD) of Big data urban collections are chosen.
According to the results of the study, a list of Big Data characteristics has been formed, which is presented in the format “10V”, containing the attributes as volume, velocity, variety, validity, value, veracity, visibility, virtual, variability, and valence. It is denoted
V BigData ( Attr BigData ) , the set of Big
Attr BigData = Attri BigData , i = 1,10 , that provides the fixed characteristics.
volume, velocity, var iety, validity, value, veracity, Attr BigData = (1) visibility, virtual, var iability, valence
For each of these characteristics, it is created the lists of parameters that can uniquely identify the presence of the i -attribute:
Pi BigData = PJBiigData , Ji = 1, Ni (2) where Ni is the number of parameters that describe the i -characteristic Attri BigData of Big Data.
Further development of the Big Data concept leads to an extension of the
presented list of characteristics [
Step 3. It is set up a list of alternatives containing a few Big Data analytical processing methods, in particular:
Managed machine learning [
Unmanaged machine learning [13].
Data mining [14].
Statistical analysis [15].
Data visualization [16].
Available alternatives to Big Data analytical processing methods form the lower level of the hierarchy. The structure of the decision-making problem by the hierarchy analysis method on the choice of the method for analytical processing of Big Data urban collection to support the processes in the resource social and communication networks is presented in Figure 2.
Big Data analytical processing volume velocity variety validity value veracity visibility virtual variability valence Managed machine learning
Unmanaged machine learning
Data mining
Statistical analysis
Data visualization
In our case, the decision-making process is to choose one of the possible alternative categories based on the constructed priority vector. The priority will be a real number that corresponds to each alternative. The highest priority alternative would be considered a decision. The priority of the alternative will be called its weight.
Step 4. It is used the determination of the expert estimation scale for the hierarchy analysis method. In this case, an expert estimation scale or the importance of pairwise comparisons was used to evaluate the advantages of one object over another with values from 1 to 9. The overall content of the estimates is presented in Table 1. Equally important The contribution of both objects is the same 3 Weak significance Expert experience and judgment give the first object a slight advantage over the second one 5 Essential or great significance Expert experience and judgment give the first object a great advantage over the second one 7 Great and obvious significance The advantage of the first object over the second one is very large, almost obvious 9 Absolute significance The evidence to the benefit of the first object is more than convincing 2, 4, 6, 8 Intermediate values between Used for situations where compromise adjacent scale values solutions are required Reciprocal If one of the above values is obtained for comparing the first object with the of given second one, then the result of comparing the second object with the first one is values reciprocal Step 5. It is to construct matrices of pairwise comparisons for each of the above evaluation criteria and to calculate the numerical characteristics of these matrices, including the consistency index, the largest eigenvalue, and the index of ratio sequence. Each of these matrices contains the value of expert estimation on pairs of alternatives, which are the methods of Big Data analytical processing from the given list. The volume characteristic is the value of Big Data urban collections [17]. A pairwise comparison matrix for selecting the category of Big Data analytical processing methods by the volume attribute is presented in Table 2.
Managed Machine Learning 1 0,5
Machine 0,3750 0,5217 0,3158 0,2353 0,2727 1,7205 Learning Unmanaged
Machine 0,1875 0,2609 0,4211 0,3529 0,3636 1,586 Learning Data Mining 0,1250 0,0652 0,1053 0,2353 0,0909 0,6217
SAtantaislytisciasl 0,1875 0,087 0,0526 0,1176 0,1818 0,6266 DatazaVtiiosnuali- 0,1250 0,0652 0,1053 0,0588 0,0909 0,4452
Total 1 1 1 1 1 5 The column elements of the Table 3 will be obtained thanks to the normalization procedure applied to the corresponding column elements of the Table 2. To estimate the principal eigenvector of a pairwise comparison matrix, it is calculated a vector of priorities whose elements are the weights of alternatives calculated in the form of algebraic sums of the elements of the corresponding rows in the Table 3, divided by the total number of alternatives, namely the number of line items in the Table 2.
Therefore, the best alternative is the Managed machine learning category by the volume criterion because it has the highest value of weight – 0,3441. For the pairwise comparison matrix constructed according to the volume criterion, the following parameters are calculated: the estimate of the largest eigenvalue, calculated by the formula:
n max = wi si (3)
i=1 where wi is the weight of the alternative with i number, Si is the sum of the column elements of the pairwise comparison matrix with i number, n is a number of alternatives; the consistency index:
CI = 1,25 = 0,0898
(7) = 0,0725
(8) CR =
I =
RI
Since CR = 7,25% 10% , the matrix of pairwise comparisons according to the volume criterion is considered to be consistent one. Pairwise comparison matrices for selecting the category of Big Data analytical tools by criteria such as velocity, variety, validity, value, veracity, visibility, virtual, variability and valence are formed similarly to Table 2. The results of the weight estimation of the alternatives for these criteria are also formed in the same way as in Table 3. The best alternatives to these criteria and the corresponding calculated values are given in Table 4. Similarly, to the volume criterion, it is estimated the largest eigenvalues – max , the consistency indices – C I and the sequence of ratios – CR , presented in Table 5. Since the inequality CR 10% is satisfied for all the criteria of the consistency ratio, then all pairwise comparison matrices are consistent.
Step 6. It is an evaluation of the importance of the criteria to determine the weight of the alternatives. The criteria are equally important to simplify the calculations. In the pairwise comparison matrix, all squares will be filled with the same values equal to one.
In the column of total value in Table 6, the alternatives of analytical processing methods are calculated in the form of arithmetic mean of weights. 0,3172 0,1243 0,1253 0,0890 0,0898 1 0,4143 0,1961 0,1232 0,0384 1 0,0662 0,0761 1 0,4987 0,0895 1 1,4178 0,7390 10 0,1418 0,0739 1 Step 7. It is to select the methods with the highest weight. According to expert estimates obtained by the method of hierarchy analysis, it is the Managed machine learning category of analytical processing tools that is considered the best option for use in the analytical procedures of Big data urban collections. Let us estimate the average consistency ratio of the hierarchy according to the weighted index of the ratio sequence:
CR =
C
I = 0,0739
= 0,0596 0,1.
RI 1,25
The result shows that the entire hierarchy is consistent. The results of the calculations are presented in Table 7 and Figure 3. 0,0882 0,0974 1
A detailed analysis of Big Data processing methods was conducted, including managed machine learning, unmanaged machine learning, data mining, statistical analysis, and data visualization. A comparative analysis of their functionality was performed using the method of hierarchy analysis based on expert estimation. Furthermore, the evaluation was performed on 10 criteria such as volume, velocity, variety, validity, value, veracity, visibility, virtual, variability and valence. According to the results of the study, the managed machine learning is determined as the most effective method. 13. Al-Turjman, F.: Smart Cities Performability, Cognition, & Security. Springer International
Publishing (2019). 14. Honarvar, A. R., Sami, A.: Towards sustainable smart city by particulate matter prediction using urban big data, excluding expensive air pollution infrastructures. Big data research, 17, pp. 56-65 (2019). 15. Soomro, K. et al.: Smart city big data analytics: An advanced review. Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery, 9.5 (2019). 16. Bouloukakis, M. et al.: Virtual Reality for Smart City Visualization and Monitoring. Mediterranean Cities and Island Communities, Springer, Cham, pp. 1-18 (2019). 17. Osman, A. M. S.: A novel big data analytics framework for smart cities. Future Generation Computer Systems, 91, pp. 620-633 (2019).