Indian Research on Artificial Neural Networks: A Bibliometric Assessment of Publications Output during 1999-2018

1. Introduction

Artificial neural networks (ANN) have been around for a long time as a subset of deep learning concepts (McCulloch & Pitts, 1943). Neural networks are designed to learn and make intelligent decisions on their own on complex tasks such as classification of images, pictures, and concepts and also address prediction problems (Lek & Park, 2008). Over time, networks have expanded in data processing capabilities and the ability to process more complex tasks. ANN algorithms simulate the processing concepts of the human brain to model complex patterns and predictions problems. ANN algorithms work to extract information from raw data and represent it in some type of model. The model so developed provides intelligence on how to infer things from other data not yet modeled. Put simply, artificial neural networks have emerged as a family of computational models that exhibit artificial intelligence effective at solving tasks such as pattern recognition, learning, pattern approximation, generalization, classification, and clustering. Neural networks have been applied in diverse fields including aerospace, automotive, banking, defense, electronics, entertainment, financial, insurance, manufacturing, medical, oil and gas, speech, securities, telecommunications, transportation, and environment (Elprocus). ANN computing as a subject is fast expanding in all its dimensions, like ANN concepts, types, applications, and models. During the last two decades, a lot of papers have been published in the domain of ANN computing. Given these developments, it is deemed appropriate and necessary that a bibliometric study is undertaken to evaluate ANN research in India in the global context and describe India’s performance in the subject in terms of its global ranking, and the institutions and researchers that are driving the research within the country.

The study is designed to examine qualitative and qualitative aspects of India’s overall research output in the area of artificial neural networks as indexed in the Scopus database during 1999-2018. The specific objectives of this study are: (i) Analyze global research in the subject in terms of publications growth and global share of top 10 most productive countries, (ii) Analyze ANN research in India in terms of publications growth, its distribution by document types, source publication types, broad subject areas, network type, and ANN application sectors, (iii) Analyze ANN research in India in terms of citation impact and describe bibliographic features of highly-cited papers, and (iv) Identify top 25 most productive organizations and authors and top 20 journals for research communications.

2. Literature Review

Bibliometric/scientometric studies related to the analysis of the “Artificial Neural Networks” research covering publications and patents are not available in large numbers in the literature on the subject, both at the national and international levels. Amongst the existing studies, Kumar (2016a) examined Artificial Neural Networks (ANNs) research in India during 1991-14 on measures like research growth, global publication share, activity index, relative citation impact, and impact factor. Kumar (2016b) also examined the conformity of Lotka’s law to authorship distribution in the field of Artificial Neural Networks research (ANNs) comprising 3411 articles and 5654 unique authors. Claude et al. (2004) studied the distribution of articles involving artificial neural networks in the fields of medicine and biology as indexed in ISI databases during 2000-2001 (1803 articles and 49 countries), with a focus on the parameters: the number of articles, the total impact factor, the ISI journal category, the source country population, and the gross domestic product.

However, in addition to bibliometric studies on ANN research per se, quite a few bibliometric studies are available on the neural network. Amongst these studies, Noyons and van Raan (1996, 1998) evaluated neural network research to identify productive countries and research institutions globally. Using bibliometric mapping technique they developed a methodology of “self-organized” structuring of scientific fields. Tijssen (1993) examined perceptions of scientific experts regarding the intellectual shape and contents (cognitive structure) of their scientific domain. Amsaveni (2016) tested the possible application of Bradford law to the Indian literature (2001-15) on neural networks, consisting of 5209 articles and 58249 citations. Ponnudurai and Priya (2011) analyzed the relative growth rate and doubling time of neural network research output at the international level using time series data from 1969 to 2007.

3. Methodology

In order to undertake a study of India’s contribution to the artificial neural network research, publications data was sourced from the Scopus database (http://www.scopus.com) covering a 20-year period 1999-2018. Single keyword “Artificial neural networks” was used in “Keyword tag” as well as in “Article Title tag” (joined by Boolean operator “OR”) simultaneously and restricted the output to period “1999-2018” in the “period tag”, to get global publication data (consisting of 84,509 records). The above-described search strategy was refined by country of publication (including India) to get publication output data on the top 10 countries. India’s publication output comprised of 8260 records. The search strategy for obtaining India’s output was further refined to get statistics on India’s output by subject, collaborating country, organization, author, and journal. Citations to publications were counted from the date of their publication till 20 November 2019. Separate search strategies were formulated to get data on various types of networks, tasks, and applications. A complete counting method, wherein every contributing author or organization covered in multiple authorship papers was fully counted and used. All authors or organizations of multi-authored papers have received equal credit in data counting and analysis. All types of publications have been used in this study.

( KEY ( “Artificial neural networks” ) OR TITLE ( “Artificial neural networks” ) ) AND PUBYEAR > 1998 AND PUBYEAR < 2019 AND ( LIMIT-TO ( AFFILCOUNTRY, “India” ) )

4. Results

4.1 Publication Growth

The global research output in the field of “Artificial Neural Networks” (ANN) in 20 years was 84,509 publications during 1999-2018, an average of 4225.4 publications per year. India accumulated 8,260 publications during the period, with an average of 413 publications per year. India registered 10.53% growth compared to 24.53% by the world. India’s absolute growth between 1999-2008 and 2008-2018 was 388.74% compared to 130.64% by the world. India contributed 9.77% share to global output in 20 years; its 10-year global publications share surged from 5.49% (1999-2008) to 11.63% (2009-2018). In the field of ANN research, India’s citation impact on a 20-year window averaged to 11.95 citations per paper (CPP), and its 10-year citation impact dropped from 25.99 CPP to 9.07 CPP during 1999-2008 to 2009-2018 (Table 1, Fig. 1).

Table 1.

Annual and Cumulative Publications on “Artificial Neural Network Research” (ANN) in India during 1999-2018

Fig. 1.

Artificial Neural Networks Research Growth- Global Vs India 1999-2018

Of the total publications, 63.85% appeared as articles, 31.69% as conference papers, 2.08% as reviews, 1.33% as book chapters, and all others less than 1.0%.

4.2 Top 10 Countries in Artificial Neural Network

In all, 163 countries participated in global Artificial Neural Network (ANN) research. The distribution of ANN research across participating countries was uneven. For instance, 57 countries published 1-10 papers, 32 published 11-70 papers, 7 published 71-100 papers, 38 published 101-600 papers, 10 published 601-1000, 15 published 1001-5000 papers and 4 published 5001-16744 papers.

The bulk of the global research output (75.15%) in the field was contributed by the top 10 most productive countries alone. China and USA are in the leadership position in the world ranking, accounting for 19.81% and 14.79% global publications share respectively. India ranks third in the world ranking with 9.77% share. The global publication share of 7 other top 10 counties has been in single-digit ranging between 0.14% and 6.82% (Table 2, Fig. 2).

Table 2.

Global Artificial Neural Network Research: Top 10 Most Productive Countries 1999-2018

Fig. 2.

Global ANN Research : Top 10 Most Productive Countries 1999-2018

4.2.1 India’s International Collaboration

During 1999-2018, India’s collaboration with 10 select countries was the largest, nearly 12% (998 papers) of its total output in ANN research (8260). Five of India’s collaborative countries have also been the world’s most productive countries - USA, UK, China, Iran, and Canada (Table 3, Fig. 2). India’s collaboration with the USA was the largest (25.55% of India’s ICP output), but with the other 9 of 10 collaborative countries, its international collaborative publications (ICP) share was in single-digit between 4.51% and 7.72%.

Table 3.

India’ International Collaboration in ANN Research during 1999-2018

4.3 Subject-Wise Distribution of India’s Research Output

In all, Artificial Neural Network (ANN) research in India intersected with 14 disciplines (as identified in Scopus database classification). Of these, engineering and computer science have been the most favored subject areas in ANN research pursuits (with 45.67% and 41.79% national publications share respectively). In the other 12 disciplines, national publications share ranged between 3.04% and 10.87%.

Research activity index in all of 14 disciplines witnessed fluctuations between 1999-2008 and 2009-2018. Compared to the world average index of 100, seven disciplines registered a significant rise in their activity index, and in 5 other disciplines, it registered a significant decline. In two other areas, the decline in the activity index was marginal. Neuroscience recorded the highest citation impact per paper of 23.40 and mathematics the least (7.23) during 2008-19 (Table 4, Fig. 3).

Table 4.

Subject-Wise Breakup of Indian Publications in Artificial Neural Network Research during 1999-2018

Fig. 3.

Global ANN Research : Activity Index Change between 1999-2008 and 2009-19

4.4 Distribution of Publications by Type of ANN Research

In all, ANN research can be classified into 10 types and these 10 types account for 19.46% share of total output by India in the subject. Of these, Feed Forward Neural Network type and Adaptive Neural Network type account for the predominant share (15.56%) of the national output in ANN research in the country. The other eight types account for a marginal 3.9% national publications share (Table 5). Nearly 10% of the national output of the country was devoted to ANN research by network architecture. Of this, Radial Basis Network type research accounted for 4.97% share and Multi-Layer Neural Network type research, 3.93% national share. Research in other network architecture type accounts for 2.22% national share (Table 6).

Table 5.

Classification of India’s Publications on ANN by Type of Network during 1999-2018

Table 6.

Classification of India’s Publications on ANN by Type of Network Architecture during 1999-2018

4.5 Distribution of ANN Publications by Applications and Task

In all, 74.41% of ANN research in India was devoted to ANN applications in 19 disciplines. Of these, ANN applications in medical science and environment science accounted for the largest 11.82% and 10.84% national share respectively during 1999-2018. ANN applications in the remaining disciplines ranged between 0.08% and 9.12% national share. Among application areas, marketing registered the highest citation impact per paper of 98.43, followed by hydrology (49.24), environment science (22.89), water resources (22.74), etc. (Table 7).

Table 7.

Distribution of ANN Research by Applications during 1999-2018

Nearly 41% of ANN research in India accounts for task-specific applications: prediction (21.25%), classification (16.14$), and pattern recognition (4.27%). Prediction task accounted for the largest number of papers (255) in environmental science, followed by materials science (206), medical sciences (168 papers), energy (139 papers), chemical engineering (138 papers), etc. Pattern recognition task accounted for the largest number of papers (65) in medical sciences, followed by materials science (15), energy (14 papers), mining (13), security and chemical engineering (9 each), agriculture (8 papers), management (7 papers), etc. Classification type task accounted for the largest number of papers (238) in medical sciences, followed by energy (75 papers), mining (70 papers), materials science (62 papers), environment science (53 papers), chemical engineering (36 papers), etc. during 1999-2018 (Table 8).

Table 8.

Classification of ANN by Task in Indian Artificial Neural Network LiIterature

4.6 India’s Top 25 Most Productive Organizations

In all, 2194 organizations participated in Indian research on “Artificial Neural Network” during 1999-2018, of which 1243 organizations published 1 paper each, 543 organizations published 2-5 papers each, 251 organizations 6-10 papers each, 133 organizations 11-20 papers each and 74 organizations 21-50 papers, 50 organizations 51-289 papers each. The productivity of the top 25 most productive organizations varied from 69 to 289 publications per organization; together they contributed 41.50% (3428) Indian publications share and 54.77% (54041) Indian citations share during 1999-2018. Their HI has been above average citations per paper (Table 9, Fig. 4).

• ∙ Nine organizations registered their publications output above their group average (137.12): Indian Institute of Technology, Kharagpur (289 papers), Indian Institute of Technology, Roorkee (264 papers), Anna University, Madras (237 papers), etc.;
• ∙ Twelve organizations registered their citations per paper and relative citation index above the group average (15.76 and 1.32) of all organizations: Indian Institute of Technology, Kanpur (33.04 and 2.76), Indian Institute of Technology, Madras (24.26 and 2.03), Indian Institute of Technology, Bombay (22.94 and 1.92), Indian Statistical Institute, Kolkata (22.7 and 1.9), Indian Institute of Technology, Roorkee (22.41 and 1.88), Indian Institute of Science, Bangalore (21.91 and 1.83), etc.

Table 9.

Scientometric Profile of Top 25 Most Productive Indian Organizations in Artificial Neural Networks during 1999-2018

Fig. 4.

ANN Research in India : Top 25 Productive Organizations 1999-2018

4.7 India’s Top 25 Most Productive Authors

6546 authors participated in Indian research on “Artificial Neural Network” during 1999-2018, of which 4451 authors published 1 paper each, 1440 authors 2-5 papers each, 109 authors 6-10 papers each, 133 authors 11-20 papers each, 74 authors 21-50 papers each and 1 author 58 papers. The research productivity of the top 25 most productive authors varied from 19 to 58 publications per author. Together they contributed 8.17% (675) global publications share and 19.13% (18872) global citations share during 2009-2018. Their detailed scientometric profile is presented in Table 10.

• ∙ Eight of the top 25 authors registered their publications output above the group average of 27.0: P. Samui (58 papers), T.N. Singh (42 papers), M.C. Deo (35 papers), G. Panda (35 papers), K.P. Sudheer(34 papers), R. Rakkiyappan and K.K. Sarma (32 papers each) and P. Chandra (30 papers);
• ∙ Ten of the top 25 authors registered their citation per paper and relative citation index above the group average (27.96 and 2.34) of all authors: G.P.S.Raghava (86.21 and 7.21), K.P. Sudheer (84.88 and 7.1), A. Jain (77.0 and 6.44), R. Rakkiyappan (45.34 and 3.79), T.N. Singh (39.62 and 3.32), G. Panda (34.57 and 2.89), M. Khandelwal (34.08 and 2.85), M.C. Deo (32.47 and 2.72), S.S. Tampa (29.62 and 2.48) and N.R.Pal (29.0 and 2.43).

Table 10.

Scientometric Profile of Top 25 Most Productive Indian Authors in Artificial Neural Network Research during 1999-2018

4.8 Medium of Research Communication

Nearly 67.0% (5534) of ANN research in India appeared in 1053 journals, 25.91% (2140) in conference proceedings, 5.63% (465) in book series, 0.98% (81) as books, 0.47% (39) in trade publications and 0.01% (1) as undefined. Of the 1059 journals (reporting 5534 articles), 802 published 1-5 papers each, 165 published 6-10 papers each, 59 published 11-20 papers each, 29 published 21-50 papers each and 4 published 51-136 papers each during 1999-2018.

The top 20 most productive journals accounted for 16.77% of total Indian output in journals (covering artificial neural networks research) during 1999-2018, The 10-year output in journals increased from 14.43% to 17.33% between 1999-2008 and 2009-2018. Neurocomputing was the topmost productive journal (with 136 papers) in reporting Indian research in the field of ANN research, followed by the International Journal of Applied Engineering Research (106 papers), Applied Soft Computing (71 papers), Neural Computing & Applications (51 papers), etc. during 1999-2018 (Table 11).

Table 11.

Top 20 Most Productive Journals in Artificial Neural Networks during 1999-2018

5. Conclusion

This paper analyzes India’s research in the domain of “Artificial neural networks” (ANN) on select bibliometric indicators covering 20-year research as published during 1999-2019. During the period, ANN research by India registered a fast 24.52% average annual growth, contributed 9.77% share to global output, averaged citation impact of 11.95 citations per paper, and registered 126 papers (1.53% share of national output) as highly cited papers. In all, 163 countries contributed to global ANN research (84505 publications). The top 10 most productive countries in the world alone accounted for 75.15% bulk share to global publications output in the subject. China and the USA are in the leadership position in the world ranking, with 19.81% and 14.79% global publications share respectively. India ranks the third most productive country in the world with a 9.77% share. The global publication share of the other 7 amongst the top 10 counties has been in single digit ranging between 0.14% and 6.82%.

Engineering and computer science have been the preferred subjects in Artificial Neural Network (ANN) Research (with 45.67% and 41.79% publications share). Amongst the type of network in ANN research, Feed Forward Neural Network contributed the largest publication share (10.18%), followed by Adaptive Weight Neural Network (5.38% share), Feed Backward Neural Network (2.54% share), etc. during 1999-2018. In terms of research applications across subjects, medical science and environment science registered the highest publication share (11.82% and 10.84% share), followed by materials science, energy, chemical engineering, and water resources (from 6.36$ to 9.12%), etc. during the period.

The distribution of India research by participating organizations is skewed. The top 25 organizations (out of total 1243) contributed 41.50% publications share and 54.77% citations share respectively during the period. Indian Institute of Technology, Kharagpur (289 papers), Indian Institute of Technology, Roorkee (264 papers), Anna University, Madras (237 papers), have been the most productive research organizations in the country. The organizations leading in terms of citation per paper and relative citation index were: Indian Institute of Technology, Kanpur (33.04 and 2.76), Indian Institute of Technology, Madras (24.26 and 2.03), Indian Institute of Technology, Bombay (22.94 and 1.92), Indian Statistical Institute, Kolkata (22.7 and 1.9), etc. The distribution of India research by participating authors is highly scattered. The top 25 authors (out of total 6546) across India contributed merely 8.17% publications share and 19.13% citations share respectively during the period. Neurocomputing and International Journal of Applied Engineering Research are the top two most popular journals in the subject that published 136 and 106 papers respectively. India published the highest numbers of highly cited papers in collaboration with the USA (11 out of 126), followed by Singapore (4 papers), China and Iran (3 papers each), and others. In all, 126 highly-cited papers received a total of 22170 citations, averaging to 175.95 citations per paper.

Conclusion - The USA and China lead the global ranking in ANN research. India is the third most productive country in the world. Besides, India registered a fast 25% growth in its national output in the subject. India’s AAN research pursuits were multidisciplinary and exploratory in nature, aimed at developing innovative technologies, new ANN applications, and ANN types. The top 25 research organizations in the country contributed high-quality ANN research as their h-index have been consistently above the national average citations per paper. Indian Institutes of Technology at Kharagpur, Roorkee, and Anna University, Madras provided leadership in ANN research in the country. Computer science and engineering science were the most preferred areas in AAN research pursuits. But to reach the top position in the world ranking India will have to immensely improve its research productivity and also improve its research quality and impact.

References

• Amsaveni, N. (2016). Assessment of Bradford Law’s of Scattering to Neural Network Literature through Bibliometric Study. International Journal of Next Generation Library and Technologies, 2(3), 1-15. http://www.ijnglt.com/files/v2i3/Amsaveni.pdf
• Claude, R., Charles-Daniel, A., Jean, A., & Jean-Francois, G. (2004). Bibliometric Overview of the Utilization of Artificial Neural Networks in Medicine and Biology. Scientometrics, 59(1), 117-130. [https://doi.org/10.1023/B:SCIE.0000013302.59845.34]
• Elprocus. (n.d.). Artificial Neural Networks (ANN) and Different Types. https://www.elprocus.com/artificial-neural-networks-ann-and-their-types, /
• Kumar, S. (2016a). Scientometric Study of Artificial Neural Networks Research in India. International Journal of Information Dissemination and Technology, 6(S1), S16-20. http://www.ijidt.com/index.php/ijidt/article/view/324
• Kumar, S. (2016b). An Evaluation of Author Productivity in Artificial Neural Networks Research in India during 1991-2014. Annals of Library and Information Studies, 63, 126-131. http://hdl.handle.net/123456789/34790
• Lek, S., & Park, Y. S. (2008). Artificial Neural Networks in Encyclopedia of Ecology. Amsterdam: Elsevier.
• McCulloch, W. S., & Pitts, W. (1943). A Logical Calculus of the Ideas Immanent in Nervous Activity. Bulletin of Mathematical Biophysics, 5, 115-133. [https://doi.org/10.1007/BF02478259]
• Noyons, E. C. M., & van Raan, A. F. J. (1996). Actor Analysis in Neural Network Research: the Position of Germany. Research Evaluation, 6(2), 133-142. [https://doi.org/10.1093/rev/6.2.133]
• Noyons, E. C. M., & van Raan, A. F. J. (1998). Monitoring Scientific Developments from a Dynamic Perspective: Self-organized Structuring to Map Neural Network Research. Journal of the American Society for Information Science January, 49(1), 68-81. [https://doi.org/10.1002/(sici)1097-4571(1998)49:1<68::aid-asi9>3.0.co;2-1]
• Ponnudurai, R., & Priya, N. (2011). Studies on Research Output of Neural Network at International level: Bibliometric analysis. International Journal of Current Research, 3(6), 229-233.
• Tijssen, R. A. (1993). Scientometric Cognitive Study of Neural Network Research: Expert Mental Maps versus Bibliometric Maps. Scientometrics, 28(1), 111-136. [https://doi.org/10.1007/BF02016288]