Strengthen of a Scientific Field in Latin America: Evolutionary Computation

There are several factors involved in the maturity of a new research field:^[1–3] a substantial scientific production and impact by a solid network of scholars, the support of graduate programs, as well as the recognition of external communities of scholars.^[4–6] The research collaboration has been considered as an essential tool to construct a scientific community and leads to the development of cognitives structures shared by all its members.^[7]

The bibliometric method has been used to measure the evolution of scientific fields. The main element of analysis is the bibliographic data on published Works:^[8] original articles, reviews, books, book chapters, proceedings, as well as the indicators generated from this analysis: impact factors, number of citations, frequency of publications, institutional statistics, etc.^[9]

This branch of computational science is oriented to the study of bio-inspired metaheuristics which are stochastic search and optimization techniques. Such properties constitute “algorithms that are supported in two concepts: exploration and exploitation of the search space”.^[10–13] The foundation of this new research field lies in algorithms with a structure close to the Darwinian principles of evolutionary biology.^[14] This scheme is supplemented with the principles of modern genetics and selectionism.^[10] The combination of these three theories is called Neo-darwinism or modern synthesis of genetics.^[15,16]

The genetic algorithm, which is also called evolutionary algorithm, was generated in the first decades of the XX century. The first works on the evolution of biological systems by computers were published in the mid XX century.^[15,16] Holland and Schwefel were pioneers in the development of new informatic models in genetics.^[17] Their work constitued the fundation of evolutionary computation in the published litearture. The applications of computational software were first developed by R.M. Friedberg.^[18] LA this field of research was initiated in the 1970 decade.^[10]

There are three basic processes in the field of evolutionary computation that were developed in the 1960-1970 decades:^[11] (i) evolutionary programming, (ii) evolution strategies, and (iii) genetic algorithms. This field of research has become a transdisciplinary approach with multiple applications. In particular, in Latin America (LA) the interest in this field started in the decade of 1980 with diverse applications: texts clustering,^[19] drinking water access,^[20] industry management,^[21] stock market prediction,^[22] scheduling,^[23] electricity supply,^[24] manufacturing,^[25] data mining.^[26]

The published production in this field of research includes the following topics: definitions, basic concepts,^[27] development of algorithms,^[28] optimization of algorithms or genetic programing,^[29] application of artificial intelligence,^[30] evolutionary optimization, multi-objectives and hiper-heuristic.^[11] The trandisciplinary relation of evolutionary computation has been also studied.^[31] However, there are very few bibliometric analysis on this research field: bussiness inteligence in the agricultural sector,^[32] data visualization, data networks, genetic programation, software developments and analysis of computation journals.^[33–35] Hoevere, as far as we know, there are no studies on the bibliometry of evolutiionary computation. This precisely the aim of the present study is to characterize the progress of this field of research through the analysis of the WoS databases. In particular, we are interested in answering the followin points: which were the first research publications in this research fiueld? when does this field reached its maturity? which are the principal research collboratrions? In the following section the respective search strategics are presented, as well as our results and conclusions.

The present study uses the method associated to the systematic review of the published literatura on the field of evolutionary computation by Latinamerican researchers using the WoS databases. This method three different types of reviews use: analysis of dominio, analysis based in theories, and analysis of methods. The first case involves five stages:^[36] estructured review, review based in frames of reference, quantitative bibliometric review, theories review, and finally a hibrid review. However, in the present work we will use the bibliometric review as an essential tool to analyze different trends in this new field of research.^[37]

In this framework, the general aim of the present work is to determine the evolution of the Latin America output in the field of evolutionary computation, dando respuesta a las preguntas: which were the first research publications in this research fiueld? when does this field reached its maturity? which are the principal research collboratrions? Figure 1 shows the flow chart followed for literature search, retrieval, organization and analysis. The model applied is SPAR-4-SLR.^[14]

We use the databases of the Web of Science (WoS) in the period 1980-2020. In particular, for characterizing the consolidation process the evolution of the Latin America output in the field of volutionary computation. The productivity and impact will correspond to the publications and citations in mainstream journals included in WoS. The publications were retrieved by matching the names of the Latin American institutions with the articles from the WoS databases. Our search strategy included the following features:

TS=(Algoritmos evolutivos or Computación evolutiva or computo evolutivo or algoritmos genéticos or Programacion Evolutiva or Estrategias Evolutivas or Programacion evolutiva or Estrategias de evolucion or Evolutionary Computing or Genetic Algorithms or Evolutionary Programming or Evolutionary Strategies or Evolutionary algorithms or Evolutionary programming or Evolution strategies) AND CU=(Argentina, or Belice or Bolivia or Brazil or Chile or Colombia or Costa Rica or Cuba or Ecuador or El Salvador or Guatemala or Jamica or Honduras or Mexico or Nicaragua or Panama or Paraguay or Peru or Puerto Rico or Republica Dominicana or Uruguay or Venezuela). Belize was considered due to its influence in terms of language and culture.

Figure 1.
Figure 1:
Diagram of flow for the search, retrieve, organization and analysis of the published literature on evolutionary computation. Model SPAR-4-SLR. Our search retrieved 16,888 documents in the period 1980-2020. The data set includes all different types of documents published in the WoS journals. The bibliometric review was completed with the data from the published production on this field of research by LA researchers.

The 16,888 documents were retrieved in blocks of 500 documents and were included in an Excel spreadsheet with normalized data: authors addresses, titles of journals, subject areas, country of the journal. We were interested in keeping extra data on each journal such as its impact factor in JCR, the respective quartile and country of edition. In particular, if the country corresponds to the LA region. Our data sample was translated to an Access file that made esay our data analysis. We needed to perform a double counting in order to determine the data on production by country and institution.

The assignment of the number of publications by continent was performed with the same methodology used for the case of LA publications: the WoS filters were used in order to label the production per year for each continent. Finally, we built the matrices required to use the VOSViewer software in the generation of network maps. Also, free software was used free for the developmen to the maps the LA collaborations with the countries of other continents.

Our results are supported by a bibliometric analysis of the published production on evolutionary computation by LA researchers: evolution of the number of articles and the respective number of citations, distribution of authors and institutions, journals and areas of knowledge, as well as scientific collaborations.

We have included in Table 1 the LA production and impact for the period 1980-2020 in the field of evolutionary computation. Our search indicates that the first LA documents in this field were published in 1980. The data included in this table shows that it was not until 1991 when the LA output started a steady trend. The number of citations per year is impressive. The average number of citations increased from 59 citations per article in 1991 to 65 citations per article in 2003 but in 2012 has reached just 27 citations per article. Of course, the latter result is expected to increase in the near future.

According to our bibliometric search, the first document published in the scientific literature was 1964. We registered just 53 new publications in the following 12 years but afterwards the respective growth dynamics became impressive as it is shown in Table . It is important to notice that in LA most of the graduate programs were funded in the 1970 decade. Our findings will show that some of these programs contributed with a substantial contribution to the research production in this field of knowledge. This was also true in other research fields.^[27]

The distribution of publications by countries of LA is included in Table 2. While Brazil has contributed with 48.4% of this output, the following four LA countries (Mexico, Argentina, Chile and Colombia) have a similar contribution (44.5%) to the whole LA output. The respective number of publishing institutions corresponds to a similar share: 41.6% in both cases. This LA trend has been observed in others research fields.^[38]

Figure 2 depicts the evolution of the scientific production in mainstream journals in the field of evolutionary computation for different geographical regions. Asia and Europe have the largest production (39.1% and 30.6%, respectively), while USA and Canada contribute with 18% of the whole production. The Latin American region contributes with just 4.9% of this production. We can appreciate also in Figure that since 2000 there a steady increase of the scientific production in this research field an all-geographical region. While Coello^[11] has pointed out that the original ideas on evolutionary computation arised in the 1950-60 decades, we can appreciate in Figure that a steady increase in the respective scientific production only started in the 2000 decade.

The LA research production in this field involves 63.2% of original papers, 3.8% in proceedings, 2.5% in review articles and book chapters. The original articles were published in 3,358 journals, of these 10.9% are regional and the rest mainstream journals. The latter, most of them in the first JCR quartiles: 35.4% in the first quartile and 31.8 in the second quartile. In Table 3 we have included the distribution of the LA production published in these journals. PLOS ONE and Astronomy and Astrophysics

Figure 2.
Evolution of the number of scientific production on evolutionary computation by geographical region (1974-2020)

have published the largest number of articles in this research field (about 200 papers each). The mainstream journals with the largest number of citations correspond to Nature, Monthly Notices of the Royal Astronomical Society and Astronomy and Astrophysics. In Table there are no regional journals because the number of published documents in this field is very low: only 11% of the total production was published in regional journals.

There is also a diversity of subject areas included in the LA production in this research field. Table 4 shows the subject areas that reached at least 1% of the whole production. The most representative subject areas correspond to Computer Science and Engineering, each with 20.9% (13.7%) and 6.6% (11.7%) of the production (citations), respectively. From the diversity of themes inlcuded in Table , we can conclude that this research field has a transdisciplinary approach with disciplines that include Ecology, Management Science, Mathematics, Physics, Chemistry, Genetics and Heridity and even Astronomy and Astrophyics.

Figure 3 depicts the collaboration network of authors for the LA production in this research field. It is easy to identify some of the leading authors in the network: Carlos Artemio Coello Coello, Oscar Castillo and Erik Cuevas from Mexico; Leandro dos Santos, Frederico G. Guimares and Ricardo Takahasi from Brazil; and Patricia Melin from Mexico. This map depicts the collaboration network of just 25 publications of the more representative authors in this research field. The color code indicates the different authors involved in the network and their connection to other authors.

We have included in Figure 4 the map corresponding to the LA collaboration with other countries. We have included for each country the number of papers published in this period, as well as the number of countries for each continent involved in the LA collaboration and the respective percentage. There is a high percentage (64%) of the number of publications published within the same LA countries. As expected, the countries with an intense LA collaboration are Brazil, Mexico, Argentina, Chile and Colombia. The collaborations with Europe and North America contribute with 19% and 11%, respectively. There is a weak collaboration with Asia (3.4%). We include in Table 5 the list of countries of each continet involved in the LA collaboration in this research field. The countries wuth the largest number of publications with LA countries are USA, Spain, England, Germany, Italy, Portugal and China.

Finally, inTable 6 we have included the LA academic instituions with the largest activity in this reserach field. Most of these institutions correspond to Brazil, Mexico, Argentina, Chile and Colombia, mainly public and privte universities, as well as research centers.^[9] The lack of specific public policies in this reserach field has not promoved the creation of newe reserach groups in this field in other LA countries.^[39]

Figure 3.
Research network of authors for the LA production in the field of evolutionary computation (1980-2020)

Figure 4.
Collaboration map by LAcountry in the field of evolutionary computation (1980-2020)

The scientific activity in LA has been scarce compared with the research production in developed countries.^[38] Some authors have emphasized that this situation involves two factors: (1) there is a low investment in science and technology in most of the LA countries, and (2) the research infrastructure is deficient with a very low number of active researchers.^[40–42] Even more, the LA governments and enterprises prefer to buy new technology instead of promoving the research activity within their own countries.^[43]

In this context, it is impressive the research activity carried out in some LA countries like Brazil, Mexico, Argentina, Chile and Colombia, in the field of evolutionary computation. Most of this activity has been developed with international collaborators^[41] with an impressive number of publications that have also generated a great deal of citations. In LA, this research field has been recognized as a frontier subject within a transdisciplinary approach.^[11] Our findings indicate that there is a steady growth of publications and citations in this research field. This output became important since the 1990s, 30 years after the first publications in this field appeared in mainstream journals. The consolidation of this research field is evident since the year 2000 with a large number of scholars and institutions active in several LA countries.^[42] The LA academic institutions, like universities and research centers, have been the most productive in subject areas like genetic algorithms, metaheuristics, neural networks, epidemiology, education research, star formation, and others.

Several public policies have boosted this area of knowledge. In particular, the Union of South American Countries (UNASUR) have contributed to improve the scientific activity in this geographical region.^[43] Some areas of knowledge have received increasing attention since 1990, in particular the social studies of science, computational sciences, and the applications of technological infrastructure.^[44] The research in computational sciences has been supported traditionally by some LA governments as a tool to improve the economic growth, as it has been the case in developed countries.^[45] Blandon has emphasized that the academic communities should make a better effort in improve the industrial progress in each country. This is particularly true in the case of evolutionary computation in the software and hardware sectors.

To summarize our results, it was found that there are very few analyses of the LA production on evolutionary computation. In the present work we have used the bibliometric method in order to determine the evolution of this research field. However, there are some other aspects of this research field that need to be analyzed such as the characterization of graduated programs. Finally, our findings indicate there is an impressive increase in the number of publications in this research field, and in our opinion a question is relevant: Is there enough infrastructure in LA in order to attend this pattern?

The authors appreciate the support from SNI and Conacyt (Mexico, research project P-A-S1-9013). The authors also thank to Professor Carlos Artemio Coello Coello for useful suggestions in the subject.