Huge amounts of data are produced every day and extracting its information is essential to predict, interpret and create various smart applications in several fields, such as science, healthcare, education, financial modeling, policy, marketing, etc. (Sarker, 2021; Thrun et al., 2021; Zheng et al., 2024). Therefore, data management tools and techniques for advanced analysis that can extract insights and useful knowledge from vast data are needed.

In recent years, the acceleration of technological progress and the increase in computing capacity has increased, giving rise to the well-known fourth industrial revolution (Bughin et al., 2017; Velarde, 2019; Borges et al., 2021). In this sense, Artificial Intelligence (AI) development and its real-world applications have gained popularity due to the high results in terms of accuracy and efficiency, even surpassing humans’ performance (Angelov et al., 2021).

Within that field, AI tools such as Machine Learning (ML) and Deep Learning (DL) algorithms have been capable of processing huge datasets efficiently, given that help to save time and maximize computing tools (Emmert-Streib, 2021; Xu et al., 2021). Sometimes the terms AI, ML, and DL are used as synonyms since they are closely related; however, it is important to distinguish the difference between them. Figure 1 shows the overview of these terms, further information can be consulted in specialized literature (Goodfellow et al., 2016; Mohammed et al., 2016; Chollet, 2021; Sarker, 2021).

Over time, there were key events that have led to the development of artificial intelligence, as shown in Figure 2. Several ML and DL algorithms were developed (e.g., Random Forest, Support Vector Machines, Neural Networks, KNN, etc.), and depending on the nature of the task, there are different approaches based on the type and volume of the data.

ML is a subfield of AI that allows computers to perform and improve tasks without explicit programming, with the development of algorithms and statistical tools (Jordan and Mitchell, 2015). In general, ML techniques can be classified into four groups (Mohammed et al., 2016; Hurwitz and Kirsch, 2018; Jhaveri et al., 2022):

Meanwhile, modern learning techniques have been coupled with Deep Learning architectures such as transfer learning and adversarial learning (Emmert-Streib, 2021). This since ML finds it difficult to the optimal combination of hyperparameters, extracted features, and pre-processing methods from a dataset. While DL approaches employ hierarchical layers to assemble levels of abstraction and model complex systems (Polson and Sokolov, 2020; Siddiqui et al., 2022).

DL is a branch of ML that applies Artificial Neural Networks (ANNs) that are characterized by numerous hidden layers. DL algorithms are commonly applied in pattern recognition systems, due to DL can be able to select optimal attributes for raw datasets (Alves de Oliveira and Bollen, 2023). DL has been applied and coped successfully with the high dimensional, noisy, and unstructured dataset (Janiesch et al., 2021). Moreover, DL has been widely applied in several areas of knowledge such as health issues (Pirovano et al., 2021), hydrological research (Chen et al., 2023), natural sciences (Garcke and Roscher, 2023), safety, road survey, and bridge inspection (Xu et al., 2021). In addition, outlier detection with Deep Learning methods such as reconstruction error, predictive error, and dissimilarity (Smejkalová et al., 2023).

In this current age of big data, ML and DL have become popular because of their learning capabilities from the past and their ability to make intelligent decisions (Sarker, 2021). Worldwide, exponential growth can be observed from several fields such as environmental science (Ardabili et al., 2020; Dokic et al., 2020; Saha et al., 2022; Ahmadi et al., 2022; Tao et al., 2022), medicine (Zhang, 2017; Valliani et al., 2019; Huang et al., 2020; Lui et al., 2020) or financial market (Warin and Stojkov, 2021; Ahmed et al., 2022). According to bibliometric studies, the United States of America (USA) and China are the leading countries in AI research, followed by Germany, the United Kingdom, India, Canada, and France (Rincon-Patino et al., 2018; Savage, 2020).

Particularly, in Mexico AI has gained attention in the last two decades, Figure 3 shows the timeline chart of some main events regarding AI in Mexico, based on the report by Villegas-Vergara et al. (2021). Most of the relevant events were the creation of research centers and societies related to AI such as Centro Nacional de Cálculo (CENAC), Mexican Society of Artificial Intelligences (SMIA), Laboratorio Nacional de Informática Avanzada, A.C. (LANIA), Centro de Investigación en Inteligencia Artificial (CIIA) and the Sociedad Mexicana de Ciencia de la Computación (SMCC). From the year 2000, international conferences in AI were presented as well as the inclusion of academic programs related to this field.

Although some developments have been carried out, the application of AI in Mexico continues in its early stages. According to the World Government AI Readiness Index 2022, published by Rogerson et al. (2022) from Oxford Insights (OI), Mexico was placed 62nd in the rankings out of 161 countries. In this report three pillars named (1) Government, (2) Technology Sector and (3) Data and Infrastructure were evaluated. In academic research areas, between the years 2002–2017, Mexico’s National Council of Humanities, Science, and Technology (CONAHCYT) supported about 144 projects related to AI, including the computer sciences, data and information sciences, electronics, and telecommunications the relevant disciplines (Martinho-Trustwell et al., 2018).

However, nowadays there is uncertainty about the academic research areas, algorithms employed, trends in publication, location, etc. about AI applications. To our knowledge, there is no systematic review that summarizes the state-of-the-art research applications of ML and DL in the country. The aim of this work is to examine the variety of ML and DL algorithms employed in Mexico country, using the PRISMA method. Under this general objective, the answers are searched for the following research questions (RQ):

To answer this question, a graph had to be constructed and analyzed. Additionally, characteristics such as publishing issues, institutions, and spatial distributions of research were explored.

To answer this question, the results of the systematic review needed to be synthesized comprehensively, thus each paper was classified into a subject area based on its aim and scope.

To search for answers to this question, the algorithms employed in each paper were identified. The results were included in a graph. Additionally, performance metrics issues were discussed.

The results obtained from this systematic review will allow us to identify gaps, challenges, and opportunities within this field. Furthermore, the information presented in this work could contribute to the planning, development, and improvement of the strategy in the national application of AI.

This systematic review provides a state-of-art of Machine Learning and Deep Learning development and its applications in Mexico in diverse fields. Through a rigorous and transparent method to minimize bias, researchers could identify gaps, trends, challenges, and opportunities, making evidence more accessible to decision-makers and guiding practice (Needleman, 2003; Paez, 2017). In this section, the detailed procedure employed in this systematic review is addressed.

Although this study was carried out by a systematic methodology, some limitations must be considered. The articles included in this review were retrieved from Web of Science and Scopus databases, through the electronic resources of the Instituto Potosino de Investigación Científica y Tecnológica A.C. Since we considered only two databases, there is a possible paper omission since other databases exist in both Spanish and English language, such as PubMed, IEEE, JSTOR, Redalyc, Scielo, among others. This study considered only journal articles to ensure that the included articles were of high quality and had undergone through a peer-review process. Technical reports, conference papers, and preprints were not included. Moreover, some manuscripts that meet the inclusion criteria were not retrieved in full text because the institution does not have full access to the content. Further reviews could also make modifications in the search string (e.g., adding the algorithm terms) to obtain more accurate results. The findings of this study provide essential information about the research panorama of AI in Mexico and its area applications, this type of work should continue to discover more gaps, challenges, and opportunities.

AI has developed sufficiently as a scientific discipline and technology, having extended from laboratories to the entire community. Nowadays, industry and government administrations are rapidly using intelligent and digital technologies in their daily tasks and undertakings, bearing in mind the Sustainable Development Goals (Vinuesa et al., 2020; Palomares et al., 2021). Research publications employing ML and DL approaches in Mexico have increased considerably in the last 5 years and are expected to continue rising, this could be an inflection point since some opportunities for development are present.

Based on the findings of this study, there is a potential to make a big scientific contribution by studying water resources using AI techniques. Worldwide, both ML and DL approaches have been applied to solve complex water-related problems, including real-time monitoring, forecasting, water resources allocation, water systems technology optimization, pollutant source identification, and pollutant concentration estimation (Huang et al., 2021; Zhu et al., 2022). Some applications in surface water included the development of water quality prediction and analysis (Wai et al., 2022; Zhu et al., 2022; Irwan et al., 2023), whereas in groundwater resources, predicted characteristics such as discharge, groundwater recharge, groundwater level fluctuation, aquifer loss coefficient, among others, has been studied by researches (Ahmadi et al., 2022). Mexico is confronted by several water difficulties, including water scarcity, pollution, and ineffective water administration. The modeling efforts mostly focused on general processes such as conflict resolution, water resources planning, water availability, and demand diagnosis, with the application of traditional software (e.g., Stella, UVQ, SWAT, MODFLOW, etc.) (Hernández-Cruz et al., 2022). In this research, a few published works related to the study of water resources were found, thus we encourage related researchers to focus on this area.

In some cases, the main limitations are the data acquisition and availability, which is a fundamental resource for AI models. Often data is inadequate and incomplete, or difficult to obtain through traditional in-situ methods. In this sense, remote sensing could provide essential information for data extraction, image classification, change detection, or accuracy assessment (Maxwell et al., 2018; Ma et al., 2019). Our research has demonstrated that published work related to remote sensing is scarce, with only 8% of the total reviewed papers; thus, there is an opportunity for development in this area.

Furthermore, where it is possible to collect high-quality data, advanced techniques such as Transformers can open the door to capturing temporal relationships in history for prediction or classification tasks, where they have been successfully applied in related works, even (Méndez et al., 2022; Maurício et al., 2023; Islam et al., 2024). Thus, we encourage you to explore this area.

Additionally, if there are no conflicts of interest or legal issues, Open-source AI models and data sharing are suggested as ways to enable rapid development and creation of new models. In Mexico, the National Digital Strategy encourages open data sharing through the https://datos.gob.mx/ platform.

Our findings in this study should not be generalized, since only provide academic scientific production. The application and use of AI tools in the country have taken place in big companies (e.g., industry, computer science, business, telecommunications) as well as in education at all levels with the implementation of new technologies (Martinho-Trustwell et al., 2018). However, toward a robust national AI strategy, transdisciplinary collaboration between academia, industry, and civil society is recommended. The results of this study could provide essential bases to continue the scientific production in the country, toward the development of guidelines for an AI strategy.

This work is the first approach to summarize the current state-of-the-art in the research panorama of ML and DL models within a developing country such as Mexico. A systematic methodology was conducted via the PRISMA 2020 statement. Summarizing the trends in publications provided the answer to RQ1. The publications in the country were scarce over a decade, having a significant increase in the last 5 years, with two peaks in 2021 and 2022. Furthermore, most of these studies have been carried out in the central zone of the country, showing a location bias. Forty-three institutions were identified with research publications, where the Universidad Autónoma Nacional de México presented more in comparison with the rest. The aim and scope of each paper allowed us to answer RQ2. A total of 15 subject areas were identified, where Social Sciences and Medicine were the main areas of applications, whereas areas such as Geosciences were less explored. Exploring the answer to RQ3 led to the identification that ANN models were preferred, probably due to their capability to learn and model non-linear and complex relationships. Other popular models included RF and SVM. In general terms, the selection and application of the algorithms rely on the study objective and the data patterns. Regarding the performance metrics applied, accuracy and recall were the most employed. This review presented a general panorama of the research publications in the AI field, this work will help readers to understand the several ML and DL techniques used and their subject area of application. Moreover, the study could provide significant knowledge in the development and implementation of a national AI strategy, according to country needs as well as encourage multidisciplinary and collaboration opportunities.