A Machine Learning Based Solution for Forecasted Economics Predicting Business Dynamics Across Europe Using Open Government Data

Abstract

In an era where data-driven decision-making is paramount, this paper explores the use of advanced statistical methods and machine learning to forecast business dynamics within a specific continent, Europe. The importance of forecasted economics is discussed in the introduction, followed by a review of the relevant literature which highlights recent work and forecasts. The work has leveraged Open Government Data (OGD) from Scotland and Wales for the period 2010 to 2023, this study also employs an Artificial Neural Network (ANN) for regression analysis to predict the growth of businesses. Detailed attention is given to the parameters of the ANN used to ensure methodological transparency. The model's efectiveness is evaluated using Mean Absolute Percent-age Error (MAPE) and the coeficient of determination ( 2), with remarkable results presented as 0.8% and 0.97, respectively. Later, results are visually represented through various techniques to the purpose of comparing predicted outcomes with actual data. The paper concludes by outlining the significant contributions of the study and it also emphasizes the enhanced capability of ANNs in economic forecasting and their potential impact on policy-making.

Keywords

• eol>Forecasted Economics
• Artificial Neural Networks
• Open Government Data
• Business Dynamics

1. Introduction

ics in Europe is crucial for predicting fu-ture economic health and planning strategic policies that promote susOver the past few decades, European economics and tainable growth. The significance of data-driven forebusinesses have experienced significant shifts marked by casting in economic decision-making has been increasperiods of growth and stagnation [1, 2, 3]. These changes ingly recognized in recent scholarly work. Rodgers et al. are evident in the fluctuating number of businesses across ( 2024 ) introduced a me-thodical approach known as the the continent, a key indicator of economic health and Business-Driven Data-Supported (BDDS) process, which future conditions. In regions like Scotland and Wales, emphasized the importance of starting with clearly dethe economic landscape has been notably influenced by ifned business problems to efectively harness Big Data both regional policies and broader European trends [4, 5]. and AI for actionable insights [9]. Their process is wellThese areas have seen a rise in businesses, driven by tech- supported by a prescriptive framework, the Da-ta-tonological ad-vancements, changing consumer behaviors, Information-Extraction-Methodology (DIEM), which has and government support through various initiatives. For demonstrated eficacy across various industries, particupolicymakers and business leaders, understanding these larly highlighted through a healthcare industry exam-ple. trends is essential, as the economic environment impacts Moreover, in terms of urban economics, Hatami et al. both macroeconomic stability and microeconomic activi- ( 2024 ) addressed the gap in economic forecasting by the ties [6, 7, 8]. An increase in businesses usually indicates development of a spatiotemporal deep learning model a strong economy but can also result in greater compe- that predicts the performance of non-business services tition and market saturation. Analyzing these patterns in small urban areas [10, 11] or for peer to peer commuthrough forecasted economics provides insight into the nications and video streaming [12, 3, 13] and other appliforces shaping these trends. Focusing on Europe, espe- cations [14, 15]. The re-search utilized LSTM networks cially Scotland and Wales, allows for a unique study of and underscored the complexities of economic forecastthese phenomena on both local and continental levels. ing at the micro-geographic level. The findings revealed This analysis helps compare dif-ferent regions and un- significant predictive accuracy in employment, business derstand the specific economic drivers and challenges sales volume, and labor productivity within Mecklenburg faced by businesses. Thus, examining business dynam- County, North Carolina. Further emphasizing the theoretical underpinnings of business dynamics, Dominko et al. (2023) conducted a bibliometric analysis focusing on the circular economy within business and economics fields [16]. The study highlighted the need for action-oriented research to expedite the transition from linear to circular ifndings from the model. Finally, Section 5 summarizes economic models which sug-gests a focus on sustainable the key con-tributions of the study and suggests avenues supply chains, waste management, and business model for further research. in-novation. Furthermore, machine learning’s role in forecasting and estimation was elabo-rated by Ahamed et al. (2023) as they evaluated various machine learn- 2. Experimental Methodology: ing algorithms for their applicability in forecasting sales Design, Materials and Methods of truck components [17]. The comparative analy-sis presented the superiority of the Random Forest Regres- Open Government Data (OGD) is a central data managesion model over others like Simple Linear Regression and ment approach used in the transparency, accountability, Ridge Regression, thus providing a robust foundation for and public engagement spectrum to facilitate the release future business forecasting endeavors. The dynamics of of data through oficial governmental portals. The prieconomic growth forecasts have also been explored in mary purpose of these portals is not only to be a data relation to business cycles, also useful for integration repository, but also a platform that enables various stakeof diferent communication networks [ 18, 19]. Huh and holders to discover and reuse the important information. Kim (2020) investigated revisions in growth forecasts and OGD is used in a growing number of re-search activities uti-lized data from the Survey of Professional Forecasters such as trend analysis, predicting economic indicators, [20]. The study found a distinct asymmetry in forecast and evaluating policy impacts because it is structured revisions across economic cycles, with more significant and contains a huge amount of data. In the context of exadjust-ments during economic contractions compared to ploring the dynamics of business growth within Scotland expansions. Lastly, Aminullah ( 2024 ) delved into fore- and Wales, the methodology adopted in a recent study casting technology innovation and economic growth in provides a relevant framework [22, 23, 24, 25, 26]. It was Indonesia by the application of a feedback economics per- reported that data were systematically collected from the spective [21, 19]. The study emphasized the inte-gration open govern-ment portals of 28 European Union counof general-purpose technologies with industrial policies tries with a focus on various parameters and indicators to ensure sustainable and inclusive growth post-COVID- over a three-year period. Likewise, but not similarly, the 19. The study also provided a unique insight into the rein- current study had used the same OGD but for a period dustrialization process and its implications for economic from 2010 till 2023, for an updated timeline. This comprepolicy. Although extensive research has been conducted hensive dataset was smoothly assembled through manual on economic forecasting with ma-chine learning, gaps retrieval from public sources and oficial OGD portals, persist, particularly in applying these technologies to pre- which are freely accessible on the web. Furthermore, dict busi-ness dynamics using open government data in to synthesize the data, density-based spatial clustering specific European regions. Previous studies have mainly techniques (DBSCAN) were ap-plied, and the cluster vafocused on broader economic indicators or non-business lidity was assessed using the Davies–Bouldin index as services, often overlooking the potential of neural net- depicted in Figure 1. Drawing from the same sources, works to utilize publicly available datasets for predict- this study will utilize the dataset compiled by the aforeing business growth. Furthermore, the unique economic mentioned references and then to apply it to forecast contexts of Scotland and Wales have seldom been the business growth through the deployment of ANN modsubject of such detailed predictive analyses. This study els. The approach ensures a robust analytical ba-sis for addresses these gaps by employing an advanced Artificial the examination of regional economic trends within the Neural Network (ANN) to analyze open government data specified timeframe. from 2010 to 2023, aiming to forecast business growth in Scotland and Wales. This approach not only deepens the understanding of regional economic dynamics but also advances the literature by providing methodological innovations in using neural networks for economic forecasting. The novelty of this work lies in its integration of detailed neural network parameterization with a focused application on open government data, ofering pivotal insights for policymakers and economic strategists. The remainder of the paper is structured as follows: Section 2 describes the data collection process and the Figure 1: Dataset creation process [22] setup of the neural network model. Section 3 delves into the specifics of the ANN configuration and the rationale behind the chosen parameters. Sec-tion 4 presents the

3. Machine Learning

Machine learning (extended from artificial intelligence) is centered around the idea that we can make computers learn based on data and make decisions without writing a specific code to accomplish this in each case [27, 28, 29, 30]. It automatically detects patterns, creates predictions, and extracts actionable insights from large datasets, changing research across various disciplines [31, 32, 33, 34, 35]. At the heart of machine learning are algorithms that improve automatically through experience and by processing data. Of these algorithms, Artificial Neural Networks (ANNs) participate by imitating the neural structure of the human brain and have Figure 2: Neural network workflow [38] the ability to estimate non-linear relationships in data [36, 37, 38, 39, 40, 41]. ANNS are modeled lightly based on the human brain; they are composed of connected Table 1 nodes, or ’neurons.’ The nodes are made of layers that ANN used parameters process the input data by various transformations. Hier- Parameter Value archy of Networks - These networks have an input layer that takes in the data, some hidden layers that process NumbeNruomfnbeeurroofnhiinddfiersnt lhaiydedresn layer 120 the data, and at the end, an output layer that gives the re- Number of neuron in second hidden layer 4 sults of the prediction or, in some cases, the classification. Solver Adam Feedforward algorithms handle large numbers of datasets Number of iterations 500 having complexity, which makes them perfect for image recognition, natural language processing, and economic forecasting. In economic terms, this provides vital infor- visual representations provide a clear depiction of trends mation in order to predict the future, which is aimed at over the period from 2010 to 2023. Figure 3 presents a influencing the sort of strategies that organizations use free visualization of the number of businesses in Scotwhen planning for the future In the implementation of land as it illustrates a general trend of growth with some the ANN for this study, key parameters and their specifi- lfuctua-tions over the examined years. Notably, a peak is cations are important for functionality and replication, observed in the years leading up to 2019, followed by a as depicted in Figure 2 and detailed in Table 1. Figure 2 il- noticeable decline, which may correlate with economic lustrates the basic structure of the ANN, where denotes or policy changes during that period, or the pandemic as the output, represents the weights associated with the we all know it. Conversely, Figure 4 fo-cuses on Wales by inputs, and indicates the bias, all processed through the showcasing of a slightly diferent trajectory. The vian activation function denoted by This schematic pro- sualization in-dicates a steady increase in the number of vides a visual representation of the data flow within the businesses up to 2020, followed by a dip likely influenced network. Table 1 outlines the specific parameters used in by external economic factors, before a modest recovery the construction of the ANN. The network comprises two in subsequent years. These visual analyses highlight hidden layers, the first one contains ten neurons and the the regional economic conditions and also pro-vide a second contains four. The Adam solver was chosen to op- comparative perspective on how diferent areas within timize the network since it is known for its eficiency in Europe are evolving in terms of business establishment handling large datasets and noisy problems. The training and growth. The data ofer a general valuable insight for process was conducted over 500 iterations which allowed regional economic planning and development strategies. suficient time for the network to converge and adjust its weights and biases to minimize prediction errors to maintain robustness in the forecasting model. 4.2. Forecasts and Regression Results In evaluating the predictive capabilities of the neural net4. Results and Discussion work model applied to the business dynamics in Scotland and Wales, the forecasted results are quantitatively as4.1. Data Visualization sessed through key performance indicators, namely the In the exploration of business growth dynamics within MAPE and R2. As detailed in Table 2, the model achieved the United Kingdom, partic-ularly in Scotland and Wales, a MAPE of 1.1% and an R2 of 0.966 for Scotland, this is ing trends in the Scottish business population. Similarly, Figure 6 is for Wales and shows that the forecasted line of best fit is closely aligned to the actual business count, meaning the initial forecast of around 193,571 closely mirrors the actual 190,800 business count in 2010 and expected 217,067 by 2023 is therefore not far from the actual 219,000 businesses. Not only does this visual representation exacerbate the statistical results, it also clearly shows just how well our model was able to adapt to the economic circumstances in Wales. The low MAPE figures and high 2 scores for these results confirm the ANN based technique to be a powerful tool for business number forecasting. It is concluded that the results of the model show that it has the potential to be an important tool when it comes to the economic analysis and policy making in regions with dynamic economic activity. an indication of a high degree of accuracy and reliability in the forecasts. Similarly, for Wales, the model demon- Figure 5: Regression line for the forecasts of number of busistrated even more precise predictions with a MAPE of nesses in Scotland 0.8% and an R2 of 0.97 which suggests that the forecasts are nearly exact replicas of the observed data.

The regression analyses that corroborate these practical implications are illustrated in Figures 5 and 6. In 5. Conclusions and Future Figure 5 the regression line is given to help determine the Directions forecasted versus non-forecasted number of businesses for Scotland, such as a forecasted confidence at 291,996 This study successfully employed an advanced ANN to catching up on an original count from 2010 of 285,000, forecast the number of businesses in Scotland and Wales and a forecast of (303,065) approaching near the original from 2010 to 2023, utilizing OGD. The ANN model demonvalue of 298,300 in 2023. The close alignment between strated high predictive accuracy, as evidenced by MAPE the predictions of the current approach and the future and the 2 values, which were notably low and high reresults provided by the administrative data model under- spectively, indicating the model’s eficacy. The re-search lines the model’s ability to capture the genuine underly- significantly contributes to the field by highlighting the

References

cluding subseries Lecture Notes in Artificial Intel- apulia region (southern italy) , Nutrients 13 ( 2021 ).

ligence and Lecture Notes in Bioinformatics) , vol- doi:10.3390/nu13093006.

ume 7267 LNAI , 2012 , p. 21 - 29 . doi: 10 .1007/ [26] V. Ponzi , S. Russo , A. Wajda , R. Brociek , C. Napoli,

978-3- 642 -29347- 4 _ 3 . Analysis pre and post covid-19 pandemic rorschach [14] F. Fiani , S. Russo , C. Napoli , An advanced solu- test data of using em algorithms and gmm mod-

tion based on machine learning for remote emdr els , in: CEUR Workshop Proceedings , volume 3360 ,

therapy , Technologies 11 ( 2023 ). doi: 10 .3390/ 2022 , p. 55 - 63 .

technologies11060172. [27] L. A. Al-Haddad , et al., Enhancing wind tunnel [15] S. Falciglia , F. Betello , S. Russo , C. Napoli , Learn- computational simulations of finite element analy-

image classification , Neurocomputing 588 ( 2024 ). Technol J 42 ( 2023 ) 1 - 9 .

doi:10 .1016/j.neucom. 2024 . 127654 . [28] L. A. Al-Haddad , A. A. Jaber , Improved uav blade [16] M. Dominko , K. Primc , R. Slabe-Erker , B. Kalar , A unbalance prediction based on machine learning

action-oriented research , Environment, Develop- Engineering 45 ( 2023 ) 463 .

ment and Sustainability 25 ( 2023 ) 5797 - 5830 . [29] L. A. Al-Haddad , A. A. Jaber , S. A. Al-Haddad , Y. M. [17] S. F. Ahamed , A. Vijayasankar , M. Thenmozhi , S. Ra- Al-Muslim, Fault diagnosis of actuator damage in

operations , The Journal of High Technology Man- computing 80 ( 2024 ) 3005 - 3024 .

agement Research 34 ( 2023 ) 100455 . [30] L. A. Al-Haddad , A. Jaber , Applications of ma[18] R. Giuliano , F. Mazzenga , E. Innocenti , A. Vizzarri, chine learning techniques for fault diagnosis of uavs

Integration of video and radio technologies for so- ( 2022 ).

cial distancing , IEEE Communications Magazine [31] G. C. Cardarilli , L. Di Nunzio , R. Fazzolari , D. Gi-

59 ( 9 ) ( 2021 ) 30 - 35 . ardino, M. Re, A. Ricci , S. Spano , An fpga-based [19] C. Napoli , G. Pappalardo , E. Tramontana, An agent- multi-agent reinforcement learning timing synchro-

driven semantical identifier using radial basis neu- nizer, Computers and Electrical Engineering 99

ral networks and reinforcement learning , in: CEUR ( 2022 ) 107749 .

Workshop Proceedings , volume 1260 , 2014 . [32] G. C. Cardarilli , L. Di Nunzio , R. Fazzolari , D. Gia[20] S. Huh , I. Kim , Growth forecast revisions over rdino , M. Matta, M. Patetta , M. Re, S. Spanò, Approx-

fessional forecasters , Economics letters 196 ( 2020 ) Telkomnika (Telecommunication Computing Elec-

109541. tronics and Control) 17 ( 2019 ) 1236 - 1241 . [21] E. Aminullah , Forecasting of technology innovation [33] G. Lo Sciuto , G. Susi, G. Cammarata, G. Capizzi , A

Forecasting and Social Change 202 ( 2024 ) 123333 . digestion process, in: 2016 International Sympo[22] S. de Juana-Espinosa , S. Luján-Mora , Open govern- sium on Power Electronics , Electrical Drives, Au-

ment data portals in the european union: A dataset tomation and Motion (SPEEDAM) , IEEE, 2016 , pp.

from 2015 to 2017, Data in brief 29 ( 2020 ) 105156 . 996 - 1003 . [23] S. de Juana-Espinosa , S. Luján-Mora , Open govern- [34] F. Bonanno , G. Capizzi, G. Lo Sciuto , C. Napoli,

logical Forecasting and Social Change 149 ( 2019 ) polaritons propagation for thin metals in openmp,

119769. in: Artificial Intelligence and Soft Computing: 13th [24] C. Napoli , C. Napoli , V. Ponzi , A. Puglisi , S. Russo, International Conference, ICAISC 2014, Zakopane,

I. E. Tibermacine , Exploiting robots as healthcare Poland, June 1-5, 2014 , Proceedings, Part I 13,

resources for epidemics management and support Springer, 2014 , pp. 22 - 33 .

caregivers, in: CEUR Workshop Proceedings , vol- [35] G. Lo Sciuto , G. Capizzi, R. Shikler , C. Napoli , Or-

ume 3686 , 2024 , p. 1 - 10 . ganic solar cells defects classification by using a [25] V. Marcotrigiano , G. D. Stingi , S. Fregnan , P. Mag- new feature extraction algorithm and an ebnn with

Journal of Intelligent Systems 36 ( 2021 ) 2443 - 2464 . [36] L. A. Al-Haddad , W. H. Alawee , A. Basem , Advanc-

learning , Computers in Biology and Medicine 169

( 2024 ) 107894 . [37] M. Y. Fattah , L. A. Al-Haddad , M. Ayasrah , A. A.

Infrastructure Geotechnology ( 2024 ) 1 - 18 . [38] L. A. Al-Haddad , A. A. Jaber , An intelligent fault

features, Drones 7 ( 2023 ) 82 . [39] L. A. Al-Haddad , L. Ibraheem , A. I. EL-Seesy , A. A.

Transportation ( 2024 ) 100155 . [40] L. A. Al-Haddad , Y. M. Al-Muslim , A. S. Hammood ,

Engineering ( 2024 ) 1 - 14 . [41] D. W. Hands , The many faces of unification and

History and Philosophy of Science 88 ( 2021 ) 209 -