Machine Learning Based Feature Importance Regarding Recycling Rates for Municipal Waste in European Countries

International Journal of Recent Innovations in Academic Research This work is licensed under a Creative Commons Attribution 4.0 International License [CC BY 4.0] E-ISSN: 2635-3040; P-ISSN: 2659-1561 Homepage: https://www.ijriar.com/ Volume-9, Issue-4, October-December-2025: 255-272 255 Research Article Machine Learning Based Feature Importance Regarding Recycling Rates for Municipal Waste in European Countries aAaron Gun-Hee Cha and *bBob Nam aSeoul International School, South Korea; bSeoul Innovations Research Institute, South Korea *Corresponding Author Email: [email protected] Received: October 30, 2025 Accepted: November 20, 2025 Published: November 27, 2025 Abstract Gradient Boost Regression reveals that real GDP is the variable that has the highest feature importance, followed by education level, total employment and life expectancy in feature importance ranking. Simultaneously, in the context of feature importance, Slovakia and Germany play a special role in predicting municipal waste recycling rates due to their contrasting positions. Maintaining negative correlations, these two countries account for a significant portion of the first principal component that governs the variation of variables including recycling rates. Moreover, the groupings of European countries in the context of PCA biplot or feature importance ranking can be compared with those by recycling rates only or by Ward’s hierarchical cluster analysis used in other studies. Partial Dependence Plots (PDP) more clarify the association of each variable with the recycling rates. Country specific dummy variables enhance model performance measured by RMSE or 𝑅2, which can be useful in facilitating integrated and well-coordinated policy design, not only at EU level but also at each national level. Keywords: Recycling Rates, Gradient Boost Regression, Feature Importance, Partial Dependence Plot, European Countries, PCA, Multiple Linear Regression. Introduction Waste Framework Directive (Directive (EU) 2018/851) mandates a 55% recycling rate for municipal waste, which is binding for all EU member states and must be achieved by 2025. More ambitious goals are: A 60% municipal waste recycling target by 2030 and a 65% municipal waste recycling target by 2035. These regulations are part of the EU's broader efforts under the Circular Economy Action Plan aimed at reducing the environmental impact of waste and fostering a more circular economy. Benefits of meeting EU recycling targets are broad from environmental protection and climate action, resource independence and job creation to improved quality of life. Simultaneously, the penalties of not meeting targets range from infringement procedures to financial disincentives. Despite these clear policy directives, substantial recycling performance gap persists among EU member states, ranging from 13.3% to 66.7% in 2022, which in turn invokes better understanding for broad socioeconomic conditions enabling efficient recycling system. In this study, economic factors such as real GDP, total employment, employment in ICT along with socio-economic factors such as education level and life expectancy are integrated as a set of features to broaden the association structures with recycling rates of member states while special attention is paid to the role of country wise heterogeneity. Literature Review A literature review of the determining factors for recycling rates in European countries highlights the multidimensional influences on recycling outcomes, especially within the context of differing economic, institutional, demographic, and policy environments. Key Economic Determinants Economic prosperity, commonly measured as real GDP per capita, consistently emerges as a strong positive factor for higher recycling rates [1-3]. Economically wealthy countries tend to have more resources to invest in recycling infrastructure and greater public engagement. A strong positive association between GDP per capita and recycling performance in the EU is reported, a finding echoed across 25 European countries and International Journal of Recent Innovations in Academic Research 256 across 27 countries from 2000 to 2019 [4-6]. This points to economic wealth as a key factor in recycling and circular performance [3]. Private investment in circular economy sectors also promotes recycling performance [1]. However, higher incomes might lead to greater waste generation, making the relationship complex [1]. Institutional and Policy Drivers Effective environmental taxation, government spending on environmental protection are significant factors [1,2]. Environmental taxes and targeted R&D funding can foster innovation and efficiency in recycling systems. Several empirical studies have consistently shown that environmental taxation correlates positively with recycling outcomes in the EU [3,4]. Based on this evidence, it is highlighted that such taxes not only encouraged cleaner production but also supported investments in recycling infrastructure [7]. Moreover, government expenditures can directly fund the development of robust recycling infrastructure, including advanced material recovery facilities and efficient collection systems, while also supporting public awareness campaigns that foster pro environmental behaviors [8]. Their efficacy is often amplified when combined with external incentives, such as monetary rewards or social influence, which are well-established predictors of recycling participation [9,10]. However, some research finds that public expenditure does not always translate into better recycling, partly due to inefficiencies or time lag before the benefits of R&D are realized [1]. Socio-Demographic and Educational Factors Urban population size and population density demonstrate complex associations with recycling rates. High urbanization may negatively impact recycling due to structural and logistical pressures in densely populated areas [1, 2]. Higher levels of educational attainment in the population correlate positively with recycling activity, indicating that more informed citizens are more likely to engage in sustainable behaviors [2,4]. Studies indicate that while urbanization and population growth contribute to higher waste generation, effective community participation in waste management, including recycling, can mitigate these effects [11]. This perspective is further supported by observations that public participation is critical for the success of waste recovery activities, highlighting the dual challenge of motivating participation while sustaining involvement [12]. In the European context, population density is revealed to significantly affect municipal waste management efficiency, while population growth is reported as a key driver of rising recycling volumes in China [13,14]. Sectoral and Technological Influences The sectoral structure of the economy-such as proportion of agriculture, industry, or services also shapes recycling outcomes. Increased agricultural intensification is linked to lower recycling rates in some contexts, while other studies find positive effects depending on the specific waste stream [2,15]. Also, technological capacity and investments in advanced recycling methods are considered pivotal for high recycling rates [16]. Additional Insights Cluster analyses show advanced recycling countries such as Germany, France, Italy, and Spain combine high recycling rates with favorable economic and institutional conditions [1]. On the other hand, countries lagging often share characteristics such as lower GDP, limited private investment, and weaker institutional frameworks [1]. Therefore, macro-level factors such as economic strength, investment in R&D and environmental sectors, educational attainment, and political commitment play crucial roles in recycling performance across European countries [3]. The complex interplay of these factors underscores the need for integrative and well-coordinated policy design at both national and EU levels. Data Most data sets are from Eurostat (2025). Dependent variable is recycling rate of municipal waste from 2015 to 2022, where some values of 2023 and 2024 are added by other sources such as European Environment Agency (EEA). Figure 1 summarizes dependent variable after classifying the values into three groups (A: above 50%, B: 30%~50%, C: below 30%). Approximately half of recycling rates of countries from 2015 to 2024 belong to Group B. International Journal of Recent Innovations in Academic Research 257 Figure 1. Recycling rate of municipal waste from 2015 to 2024: A: above 50%, B: 30%~50%, C: below 30%. Summary statistics of features related with recycling rate and thus selected in this study are given in Table 1. Except for real GDP per capita, other features are somewhat indirectly associated with the dependent variable, some of which might be affected by dependent variable. ICT employment is included as it is possible that technological innovation and digitalization can be a driving force in the waste management sectors [17]. Tertiary education can be positively correlated with the recycling rate and thus selected as an explanatory variable [18]. Citizens of countries with higher life expectancy can be more concerned with environmental issues and thus life expectancy is added [19]. Higher recycling rates can reduce domestic net greenhouse gas emissions, suggesting the possibility of negative association [20]. Although straightforward relation is vague, output of the agricultural industry can indicate economic development level, which in turn is closely related with recycling rate of European countries [21]. In 2020, about 1.8 million people were employed in municipal waste recycling and related activities [22], suggesting that total employment and recycling rate can be positively associated. Nominal labor productivity can be a proxy for economic development and technological capacity and thus can be positively associated with the recycling rate [1]. Many other features that are already assessed as significant are not included in this study, while the features that are vague in their link with recycling rate are intentionally added to broaden the scope of possible explanatory variables. Table 1. Summary statistics of explanatory variables. Figure 2. Recycling rate with year. Figure 3. Recycling rate with life expectancy. International Journal of Recent Innovations in Academic Research 258 Line graph of Figure 2 reveals clearly increasing pattern of recycling rate as time passes. The steep growth after 2023 might be related with a measurement error because the values in this range is substituted from other sources to fill the gap in Eurostat. Figure 3 of line plot shows overall increasing pattern but fluctuation is very high. Figure 4. Recycling rate with education level. Figure 5. Recycling rate with ICT employment. Line plot of education levels measured by the population of educational attainment exhibits increasing pattern after some threshold of 12000, whereas the pattern is very unpredictable below this level. In Figure 5, association is more unclear, especially below the threshold level of 1000. Figure 6. Recycling rate with real GDP. Figure 7. Recycling rate with greenhouse gas emission. Overall upward sloping in Figure 6 and downward sloping in Figure 7 after some threshold levels can be observed but the pattern is ambiguous before these turning points. Figure 8. Recycling rate with agriculture output. Figure 9. Recycling rate with total employment. International Journal of Recent Innovations in Academic Research 259 Similar characteristics such as high volatility below certain level and slight increase after that level are repeated in Figure 8, 9 which can in turn contribute to the formation of country clusters or impair the significancy of these variables. In case of nominal labor productivity, the relationship is more complex. Within the range of high volatility, tendency is vague. Beyond this volatile region, both increasing tendency or decreasing tendency is possible given different starting point. Figure 10. Recycling rate with labor productivity. Figure 11. Correlation heatmap of 10 features. Correlation heat map without country wise dummy variables in Figure 11 shows that relatively higher positive and negative correlation are mingled. Figure 12. Correlation heatmap of 10 features with country wise dummy variables. International Journal of Recent Innovations in Academic Research 260 Similar pattern is repeated in Figure 12 with country wise dummy variables. As is expected, the distribution of correlations varies by country. To more closely look into the interplay, recycling rates are grouped into three categories, integrated in segmented bar chart of each variable from Figure 13 to Figure 30. Figure 13. Life expectancy bar chart (i). Figure 14. Life expectancy bar chart (ii). Figure 14 visualizes that the proportion of higher recycling rates increases as life expectancy rises, as is expected since life expectancy is positively associated with economic development level. Figure 15. Year bar chart (i). Figure 16. Year bar chart (ii). The proportion of low recycling rates decreases while that of high rates increases over time. The frequencies of the values of 2023 and 2024 is due to data gap in Eurostat. Figure 17. ICT employment bar chart (i). Figure 18. ICT employment bar chart (ii). International Journal of Recent Innovations in Academic Research 261 Interestingly, the percent of low recycling rates is almost zero when the number of ICT employment is above 300 or the population of educational attainment is above 15k according to Figure 18 and 20. Figure 19. Education level bar chart (i). Figure 20. Education level bar chart (ii). Figure 21. Real GDP bar chart (i). Figure 22. Real GDP bar chart (ii). Real GDP is a key variable most widely investigated. Figure 21 and 22 shows that as real GDP per capita increases, the portion of high recycling rates increases while that of low rates decreases. Figure 23. Bar chart of greenhouse gas emission (i). Figure 24. Bar chart of greenhouse gas emission (ii). The countries with high recycling rates are placed in the range of greenhouse gas emissions from 4 to 13 only, while those with low rates are clustered above 30 tones. As the output of agricultural industry increases above 25k, the countries with high recycling rates almost disappear as in Figure 25 and 26. International Journal of Recent Innovations in Academic Research 262 Figure 25. Bar chart of agricultural output (i). Figure 26. Bar chart of agricultural output (ii). Figure 27. Total employment bar chart (i). Figure 28. Total employment bar chart (ii). Likewise, the portion of countries with lower recycling rates becomes almost zero when the total employment increases above the threshold level of 10k. Figure 29. Labor productivity bar chart (i). Figure 30. Labor productivity bar chart (ii). When nominal labor productivity is above 120, the countries with low recycling rates disappear. However, positive correlation between recycling rates and labor productivity is unclear because the Group B dominates above the value of 170. First Data Inspection: Multiple Linear Regression Model As the first data inspection, multiple linear regression model without country wise dummy variables is applied. Since adjusted 𝑹𝟐 is only 0.424, some key variables seem to be missing. The sign life expectancy is unexpected, indicating that overall goodness of fit is not satisfactory. Resulting regression line is given below with the output table and residual plot added in appendix as Table 8, 9 and Figure 44. International Journal of Recent Innovations in Academic Research 263 Recycling rate = 90.828(25.213) - 0.814(0.327)*Life_expectancy + 0.043(0.009)*Employment_ICT + 3.436∗ 10−4(4.367∗10−5)*RGDP - 0.295(0.129)*Greenhouse_gas_emissions - 1.723∗10−4(8.519∗10−5)*Agricuture_output – 7.719∗10−4(3.741∗10−4)*Employment_total Country wise dummy variables can help to capture the heterogeneity of countries, enhancing model performance measured by adjusted 𝑅2, which is jumped into 0.920 from 0.424, along with the significant drop of RMSE from 11.190 to 4.162. More importantly, the set of significant explanatory variables are remarkably changed. The resulting regression line with country dummy variables is given as: Recycling rate = −1548.809(205.653) + 0.803(0.102)*Year - 0.002(2.648∗10−4)*Education + 1.100∗ 10−4(2.868∗10−5)*RGDP – 0.179(0.019)*Labor_productivity + 14.281(1.726)*Austria + 14.709(1.975)*Belgium - 24.9092.015 *Bulgaria - 28.869(1.961) *Croatia - 43.041(1.944) *Cyprus + 64.578(9.912) *France - 9.765(1.901) *Czechia - 29.423(1.973) *Estonia - 9.619(1.565)*Finland + 116.115(13.153)*Germany - 30.551(1.935) *Greece15.107(1.919) *Hungary - 32.703(1.874)*Iceland + 70.722(9.225)*Italy - 23.155(2.004)*Latvia - 10.716(1.901) *Lithuania - 45.111(1.917) *Malta + 19.723(2.525)*Netherland6.244(1.602)*Norway + 25.215(5.593) *Poland - 18.841(1.889)*Portugal - 23.430(2.933) *Roma16.049(1.806)*Slovakia + 41.201(7.376)*Spain The signs of coefficients of variables such as education and labor productivity are still unpredicted, which comes from the gap between groupings of recycling rates as in the segmented bar chart and raw values in one route or from possible nonlinearity in the other route. Still possible one more aspect is that there can exist time lags in case of the education and nominal labor productivity. The population size of educational attainment does not directly indicate same year community awareness of environmental protections. Moreover, the speculation that nominal labor productivity is related with the same year technological capacity for recycling is not substantiated. In case of nominal labor productivity, this ambiguous relation is already observed in line plot and segmented bar chart. Therefore, further investigations are required to more accurately capture the relations. Second Data Investigation: Factor Analysis Below Figure 31 is a biplot from principal component analysis (PCA). The clustering of features such as ICT employment, agricultural output, total employment and education is positively correlated with the recycling rate, which forms the main part of the first principal component. Real GDP, labor productivity and life expectancy also have positive correlation with recycling rates, while greenhouse gas emissions are almost uncorrelated with recycling rates, all of which make up the second principal component. The relative contributions to the first component by ICT employment, agricultural output, total employment and education are high and similar (0.93~0.97), contrasting to the smallest effect of year (0.05~0.1) in terms of the overall variation of variables. Recycling rate is linked to both of two components. Real GDP forms the main part of the second principal component. Component loadings of these 10 variables are given in Table 2. Table 2. Component loadings of PCA. Variable Principal component 1 Principal component 2 Year 0.1074149 0.05391665 Life expectancy 0.36019388 0.70097293 Employment_ICT 0.97212441 -0.13229856 Education 0.95603946 -0.23925003 RGDP 0.15398861 0.93816334 Greenhouse gas emission -0.10073901 0.46330619 Agricultural output 0.93850353 -0.17619556 Employment total 0.96341804 -0.21425861 Labor productivity 0.18991947 0.88190638 Recycling rate 0.5011269 0.41019944 International Journal of Recent Innovations in Academic Research 270 Figure 44. Residual plot of MLR without dummies. Table 10. Output table of multiple linear regression with country wise dummy variables (i). Table 11. Output table of multiple linear regression with country wise dummy variables (ii). International Journal of Recent Innovations in Academic Research 271 Figure 45. Residual plot of MLR with dummies. Declarations Acknowledgments: Authors are thankful to the entire staff of Seoul Innovations Research Institute for insightful ideas and invaluable proofreading. Author Contributions: AGHC: Design of study, definition of intellectual content, review manuscript, data collection, prepared first draft of manuscript, data analysis, statistical analysis; BN: Statistical interpretation, prepared first draft of manuscript, review manuscript, editing. Conflict of Interest: The authors declare no conflict of interest. Consent to Publish: The authors agree to publish the paper in International Journal of Recent Innovations in Academic Research. Data Availability Statement: The datasets generated and/or analyzed during this study are not publicly available but are available from the corresponding author upon reasonable request. Funding: This research received no external funding. Institutional Review Board Statement: Not applicable. Informed Consent Statement: Not applicable. Research Content: The research content of this manuscript is original and has not been published elsewhere. References 1. Cehlár, M., Taušová, M., Ivanková, V. and Khouri, S. 2025. Municipal waste recycling in the EU: A multimethod analysis of determinants and country profiles (2005–2023). Frontiers in Environmental Science, 13: 1670365 2. Hondroyiannis, G., Sardianou, E., Nikou, V., Evangelinos, K. and Nikolaou, I. 2024. Recycling rate performance and socioeconomic determinants: Evidence from aggregate and regional data across European Union countries. Journal of Cleaner Production, 434: 139877. 3. Kostakis, I. and Tsagarakis, K.P. 2022. Social and economic determinants of materials recycling and circularity in Europe: An empirical investigation. The Annals of Regional Science, 68(2): 263-281. 4. Gabor, M.R., López–Malest, A. and Panait, M.C. 2023. The transition journey of EU vs. Non-EU countries for waste management. Environmental Science and Pollution Research, 30(21): 60326-60342. 5. Georgescu, I., Kinnunen, J. and Androniceanu, A.M. 2022. Empirical evidence on circular economy and economic development in Europe: A panel approach. Journal of Business Economics and Management (JBEM), 23(1): 199-217. 6. Holmen, R.B., Carvelli, G., Razminienė, K. and Tvaronavičienė, M. 2025. Macroeconomic influences on recycling in Europe: An econometric investigation. Circular Economy and Sustainability, 5(1): 573-602. 7. Imran, M., Jijian, Z., Sharif, A. and Magazzino, C. 2024. Evolving waste management: The impact of environmental technology, taxes, and carbon emissions on incineration in EU countries. Journal of Environmental Management, 364: 121440. 8. Chierrito-Arruda, E., Rosa, A.L.M., Paccola, E.A.D.S., Macuch, R.D.S. and Grossi-Milani, R. 2018. Proenvironmental behavior and recycling: Literature review and policy considerations. Ambiente and Sociedade, 21: e02093. International Journal of Recent Innovations in Academic Research 272 9. Hornik, J., Cherian, J. and Madansky, M. 1995. Determinants of recycling behavior: A synthesis of research results. The Journal of Socio-Economics, 24(1): 105-127. 10. Nikiema, J. and Asiedu, Z. 2022. A review of the cost and effectiveness of solutions to address plastic pollution. Environmental Science and Pollution Research, 29(17): 24547-24573. 11. Santoso, A.N. and Farizal. 2019. Community participation in household waste management: An exploratory study in Indonesia. In: E3S Web of Conferences (Vol. 125, p. 07013). EDP Sciences. 12. Oh, J. and Hettiarachchi, H. 2020. Collective action in waste management: A comparative study of recycling and recovery initiatives from Brazil, Indonesia, and Nigeria using the institutional analysis and development framework. Recycling, 5(1): 4. 13. Osińska, M. 2024. The determinants of municipal solid waste management efficiency in EU countries. Economics and Environment, 88(1): 637. 14. Huang, Q., Chen, G., Wang, Y., Xu, L. and Chen, W.Q. 2020. Identifying the socioeconomic drivers of solid waste recycling in China for the period 2005–2017. Science of the Total Environment, 725: 138137. 15. Peng, X., Jiang, Y., Chen, Z., Osman, A.I., Farghali, M., Rooney, D.W. and Yap, P.S. 2023. Recycling municipal, agricultural and industrial waste into energy, fertilizers, food and construction materials, and economic feasibility: A review. Environmental Chemistry Letters, 21(2): 765-801. 16. Önder, H. 2018. The socio-economic determiners of recycling: An analysis on European countries through a macro perspective. Amfiteatru Economic, 48(20): 405-417. 17. European Environment Agency. 2021, January 27. Digital technologies will deliver more efficient waste management in Europe (Briefing No. 26/2020). European Environment Agency. https://doi.org/10.2800/297122 18. Pela u, C. and Chinie, A.C. 2018. Econometric model for measuring the impact of the education level of the population on the recycling rate in circular economy. Amfiteatru Economic, 20(48): 340–355. 19. European Environment Agency. 2024, December 21. Waste recycling in Europe. https://www.eea.europa.eu/en/analysis/indicators/waste-recycling-in-europe 20. Eurostat. 2020, January 23. Greenhouse gas emissions from waste. European Commission. https://ec.europa.eu/eurostat/web/products-eurostat-news/-/ddn-20200123-1 21. Prakash, C., Roy, A. and Gawdiya, S. 2022. Utilization of municipal wastes in agriculture. Food and Scientific Reports, 3(6): 13-16. 22. Morgan, J. and Mitchell, P. 2015. Employment and the circular economy: Job creation in a more resource efficient Britain. Green Alliance: London. https://doi.org/10.13140/RG.2.1.1026.5049 23. Ruder, S. 2016. An overview of gradient descent optimization algorithms. arXiv. https://arxiv.org/abs/1609.04747 Citation: Aaron Gun-Hee Cha and Bob Nam. 2025. Machine Learning Based Feature Importance Regarding Recycling Rates for Municipal Waste in European Countries. International Journal of Recent Innovations in Academic Research, 9(4): 255-272. Copyright: ©2025 Aaron Gun-Hee Cha and Bob Nam. This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.