Application of machine learning to assess people’s perception of household energy in the developing world: A case of Nepal

Sep 17, 2026

Uploaded by Ivy

Research on social aspects of energy and machine learning (ML) remains limited compared with “hard” disciplines. This multidisciplinary study integrates energy, social science, and ML by comparing seven ML models for household energy classification and explaining people’s perceptions using the best-performing approach. A cross-sectional survey of 323 households in Nepal extracts 14 predictors and evaluates model performance via confusion matrix and multiple accuracy metrics. Results show Extra Trees Classifier yields the highest balanced accuracy (0.79, 0.95, 0.68), and economic constraints and low awareness shape resistance, while urban distrust sustains fuel-stacking.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages20

Reading Time50 min

AccessFree to download

Highlights

Data-driven ML model performance

Socio-economic drivers of energy behavior change

Fuel-stacking and policy implications

1. Introduction

Uncertainty in social behavior and technology adoption

Energy’s multi-disciplinary and multi-scale role

Review of social energy studies and policy awareness

Energy and AI 14 (2023) 100303 Contents lists available at ScienceDirect Energy and AI journal [homepage:](homepage: www.sciencedirect.com/journal/energy-and-ai)[ www.sciencedirect.com/journal/energy-and-ai](homepage: www.sciencedirect.com/journal/energy-and-ai) | Application of machine learning to assess people’s perception of household energy in the developing world: A case of Nepal
Utsav Bhattaraia, b, *, Tek Marasenia, c, Laxmi Prasad Devkota b, d, Armando Apana, e, f
a Institute for Life Sciences and the Environment (ILSE), University of Southern Queensland, Toowoomba, Queensland 4350, Australia b Water Modeling Solutions Pvt. Ltd. (WMS), Kathmandu, Nepal
c Centre for Sustainable Agricultural Systems (CSAS), University of Southern Queensland, Toowoomba, Queensland 4350, Australia d Nepal Academy of Science and Technology (NAST), Kathmandu, Nepal
e School of Surveying and Built Environment, University of Southern Queensland, Toowoomba, Queensland 4350, Australia f Institute of Environmental Science and Meteorology, University of the Philippines Diliman, Quezon City 1101, Philippines | | | --- | --- | | H I G H L I G H T S
• Data driven ML models are better in classifying people’s energy perceptions.
• Economy and lack of awareness contribute most to resist energy behaviour changes.
• Fuel-stacking is prevalent due to distrust in the state.
• Grass-root level responses have strong policy implications.
A R T I C L E I N F O
Keywords:
Energy
Machine learning People’s perception Socio-economy Households Nepal | G R A P H I C A L A B S T R A C T
A B S T R A C T
Research on social aspects of energy and those applying machine learning (ML) is limited compared to the ‘hard’disciplines such as science and engineering. We aim to contribute to this niche through this multidisciplinary study integrating energy, social science and ML. Specifically, we aim: (i) to compare the applicability of different ML models in household (HH) energy; and (ii) to explain people’s perception of HH energy using the most appropriate model. We carried out cross-sectional survey of 323 HHs in a developing country (Nepal) and extracted 14 predictor variables and one response variable. We tested the performance of seven ML models: KNearest Neighbors (KNN), Multi-Layer Perceptron (MLP), Extra Trees Classifier (ETC), Random Forest (RF), Ridge Classifier (RC), Multinomial Regression–Logit (MR-L) and Probit (MR-P) in classifying people’s responses. The models were evaluated against six metrics (confusion matrix, precision, f1 score, recall, balanced accuracy and overall accuracy). In this study, ETC outperformed all other models demonstrating a balanced accuracy of 0.79, 0.95 and 0.68 respectively for the Agree, Neutral and Disagree response categories. Results showed that, compared to conventional statistical models, data driven ML models are better in classifying people’s perceptions. It was seen that the majority of the surveyed people from rural (68%) and semi-urban areas (67%) tend to resist energy changes due to economic constraints and lack of awareness. Interestingly, most (73%) of the urban residents are open to changes, but still resort to fuel-stacking because of distrust in the state. These grass-root level responses have strong policy implications. | * Corresponding author. E-mail [address:](address: Utsav.Bhattarai@unisq.edu.au)[ Utsav.Bhattarai@unisq.edu.au](address: Utsav.Bhattarai@unisq.edu.au) (U. Bhattarai). [https://doi.org/10.1016/j.egyai.2023.100303](https://doi.org/10.1016/j.egyai.2023.100303) Available online 21 September 2023 2666-5468/© 2023 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license ([http://creativecommons.org/licenses/by](http://creativecommons.org/licenses/by)nc-nd/4.0/). 1. Introduction Scholars have recommended against “assuming” social behaviour, particularly related to perception and adoption of new techno

FAQs

What is the main objective of the study?

To compare seven ML models for household energy classification and use the best model to explain people’s perceptions of household energy in Nepal.

How was the dataset collected and what variables were used?

A cross-sectional survey of 323 households in Nepal was conducted, extracting 14 predictor variables and one response variable.

Which ML model performed best and how was it evaluated?

The Extra Trees Classifier outperformed other models, evaluated using metrics including confusion matrix, precision, f1 score, recall, balanced accuracy, and overall accuracy.

Sep 17, 2026

Uploaded by Ivy