Predicting Energy Poverty in Greece Using Neural Networks

In an article published in the journal Energies, researchers used artificial neural networks (ANNs) to predict energy poverty in Greece, addressing the limitations of conventional statistical models. By analyzing socio-economic and geographical factors, the authors tested three models of multilayer perceptrons, achieving accuracy levels between 61.71% and 82.72%. The results suggested that ANNs could enhance understanding and policy targeting for energy poverty.

Study: Predicting Energy Poverty in Greece Using Neural Networks.  Image Credit: Photo smile/Shutterstock
Study: Predicting Energy Poverty in Greece Using Neural Networks. Image Credit: Photo smile/Shutterstock

Background

Energy poverty, a significant socio-economic issue, deprives individuals of essential energy services. In Europe and particularly Greece, it primarily pertains to the affordability of energy, worsened by financial crises and geopolitical instability.

Previous research on energy poverty in Greece has largely relied on conventional statistical models, which assume linear relationships and thus struggle with accurate classification and prediction. For example, studies using logistic regression models highlighted a high prevalence of energy poverty but lacked precision in predictions due to their inherent limitations.

This paper addressed these gaps by employing ANNs, specifically multilayer perceptrons, to predict energy poverty in Greece based on socio-economic and geographical factors. Unlike traditional methods, ANNs could handle complex, non-linear relationships between variables, offering enhanced predictive power. The study tested three models with different input variable combinations for seven energy poverty indicators, achieving high accuracy rates of 61.71% to 82.72%, demonstrating ANNs' potential to inform more targeted policies.

Data Collection and Analytical Methodology

The researchers established a database using five primary surveys on energy poverty in Greece, focusing on different regions including the entire country, mountainous areas, Metsovo, and Agrafa. Conducted between 2016 and 2020, these surveys gathered household-level data on living conditions, energy expenses, and socio-demographic information. The database contained 1,754 data series and was used to predict energy poverty indicators through an ANN. Seven indicators, both objective and subjective were analyzed.

Data was processed using the Waikato environment for knowledge analysis (WEKA) tool, which employed a multilayer perceptron neural network. Three models were tested for each indicator, incorporating variables like house age, ownership status, household size, area, and elevation. The dataset was divided into training and test sets, 70% and 30%, respectively, and models were optimized for hidden layers and nodes. Techniques such as the synthetic minority oversampling technique (SMOTE) were used to handle imbalanced classes.

Model performance was evaluated using metrics like accuracy, precision, recall, f-measure, and receiver operating characteristic (ROC) area, alongside confusion matrices to detail correct and incorrect predictions. The research aimed not to calculate energy poverty rates but to assess the ANN's ability to predict them based on socio-economic and geographical variables, potentially enabling forecasts for any population group.

Performance Evaluation and Comparative Analysis

The researchers evaluated three ANN models, namely, models A, B, and C for predicting various energy poverty indicators using socio-geographical variables. For the "10% actual" indicator, where the household was deemed energy-poor if it spent more than 10% of its disposable income on its annual energy expenses, model C outperformed models A and B with an accuracy of 69.29%, significantly improving predictive metrics such as F-measure and ROC area. This indicated Model C's ability to predict both energy-poor and non-energy-poor households effectively.

Similarly, for the "10% required" indicator, where the household was considered energy-poor if it needed to spend more than 10% of its disposable income on its theoretically required annual energy expenses, model C achieved the highest accuracy of 81.03%, demonstrating superior metrics like precision, recall, and ROC area. This model accurately predicted energy poverty levels based on socio-geographical factors including house age, ownership status, household size, house area, and elevation.

For the compression of energy needs (CEN) indicator, where the household was considered energy-poor if it spent on energy less than 80% of the amount required to cover its energy needs sufficiently, all models performed well with accuracy scores of around 73%, suggesting reliable predictions of energy poverty.

Model C consistently outperformed across various indicators, including the National Energy Poverty Index (NEPI) and the "IW" indicator, which measured the inability to keep a home adequately warm. It demonstrated significant improvements in predictive accuracy and metric values compared to Models A and B. However, Models A and B demonstrated less satisfactory results for some indicators.

Overall, Model C's neural network architecture with one hidden layer of seven nodes, enhanced by SMOTE for class balancing, proved effective in predicting various facets of energy poverty using socio-geographical data. Compared to traditional models like logistic regression and decision trees, ANN models in this study consistently offered higher predictive accuracy, making them preferable for energy poverty assessment.

Conclusion

In conclusion, the researchers demonstrated the efficacy of ANNs, specifically multilayer perceptrons, in predicting energy poverty in Greece with high accuracy ranging from 61.71% to 82.72%. Model C consistently outperformed models A and B across all indicators, leveraging socio-economic and geographical variables.

These findings underscored ANNs' potential to inform targeted policy interventions for mitigating energy poverty, offering a robust alternative to conventional statistical models in addressing complex socio-economic issues. Future research could explore additional variables and machine learning techniques to further enhance predictive capabilities in this domain.

Journal reference:
  • Papada, L., & Kaliampakos, D. (2024). Artificial Neural Networks as a Tool to Understand Complex Energy Poverty Relationships: The Case of Greece. Energies17(13), 3163. DOI: 10.3390/en17133163, https://www.mdpi.com/1996-1073/17/13/3163
Soham Nandi

Written by

Soham Nandi

Soham Nandi is a technical writer based in Memari, India. His academic background is in Computer Science Engineering, specializing in Artificial Intelligence and Machine learning. He has extensive experience in Data Analytics, Machine Learning, and Python. He has worked on group projects that required the implementation of Computer Vision, Image Classification, and App Development.

Citations

Please use one of the following formats to cite this article in your essay, paper or report:

  • APA

    Nandi, Soham. (2024, July 09). Predicting Energy Poverty in Greece Using Neural Networks. AZoAi. Retrieved on July 17, 2024 from https://www.azoai.com/news/20240709/Predicting-Energy-Poverty-in-Greece-Using-Neural-Networks.aspx.

  • MLA

    Nandi, Soham. "Predicting Energy Poverty in Greece Using Neural Networks". AZoAi. 17 July 2024. <https://www.azoai.com/news/20240709/Predicting-Energy-Poverty-in-Greece-Using-Neural-Networks.aspx>.

  • Chicago

    Nandi, Soham. "Predicting Energy Poverty in Greece Using Neural Networks". AZoAi. https://www.azoai.com/news/20240709/Predicting-Energy-Poverty-in-Greece-Using-Neural-Networks.aspx. (accessed July 17, 2024).

  • Harvard

    Nandi, Soham. 2024. Predicting Energy Poverty in Greece Using Neural Networks. AZoAi, viewed 17 July 2024, https://www.azoai.com/news/20240709/Predicting-Energy-Poverty-in-Greece-Using-Neural-Networks.aspx.

Comments

The opinions expressed here are the views of the writer and do not necessarily reflect the views and opinions of AZoAi.
Post a new comment
Post

While we only use edited and approved content for Azthena answers, it may on occasions provide incorrect responses. Please confirm any data provided with the related suppliers or authors. We do not provide medical advice, if you search for medical information you must always consult a medical professional before acting on any information provided.

Your questions, but not your email details will be shared with OpenAI and retained for 30 days in accordance with their privacy principles.

Please do not ask questions that use sensitive or confidential information.

Read the full Terms & Conditions.

You might also like...
Dynamic Neural Codebooks Boost Vector Quantization