PaperlayerView exact revision on arXiv

Source-linked AI summary

Residential Electricity Consumption Dataset for Sri Lanka (RECON-SL)

Chanuka Algama, Merl Chandana, Isuruni Fernando, Dinithi Dissanayake, Amanda Ariyaratne, Nipuni Habaragamuwa, Pasindu Ranage, Jesús Ramos, Kasun Amarasinghe

arXiv:2609.28783v1cs.DL

TL;DR

In Sri Lanka, relatively low residential smart-meter penetration makes linked high-frequency consumption and household-survey evidence valuable. RECON-SL links longitudinal surveys with smart-meter data, and 89.3% of days in its 6-hour dataset have records from at least half of households.

  • Problem

    In Sri Lanka, relatively low residential smart-meter penetration makes evidence linking high-frequency electricity consumption with household-level survey data valuable.

    Source: BACKGROUND · p. 2

  • Method

    The paper constructs a linked resource combining high-frequency smart-meter data with three longitudinal household survey waves covering household characteristics and energy behaviors.

    Source: Survey Data Collection · p. 4, B. Objectives · p. 9

  • Results

    89.3% of days in the 6-hour dataset have records from at least half of households.

    Source: Completeness of Meter Data · p. 4

  • Takeaways & Limitations

    The dataset supports load forecasting across 15-minute-to-monthly scales and appliance-ownership prediction using combined smart-meter and survey data.

    Source: Suitable and Potential Use Cases · p. 4

    The sample covers only LECO’s service area, primarily western coastal Sri Lanka, so findings should not be generalized beyond LECO households in those regions.

    Source: Unsuitable and Cautionary Use Cases · p. 7

Abstract

from arXiv · show

This article describes the Residential Electricity Consumption Dataset for Sri Lanka (RECON-SL), a multi-source resource that integrates utility records, high-frequency smart meter data, and household surveys from the service area of Lanka Electricity Company (LECO) in Sri Lanka. The dataset captures electricity use for 4,063 households, collected over the period October 2022 to January 2025, with coverage at monthly, 6-hour, and 15-minute intervals. Monthly billing-cycle data provide complete coverage across all households, while smart meter data are available for a subsample of 1,438 households, offering both fine-grained 15-minute readings and coarser 6-hour records. Three survey waves complement these consumption measures, collecting information on demographics, housing, appliance ownership, and perceptions of energy security. Across its components, the dataset includes more than 50 million meter readings and detailed survey responses from over 11,000 household interviews. As the first dataset of its kind in Sri Lanka, and among the few worldwide to link smart meter records with longitudinal survey data at this scale, it provides a unique resource for energy, machine learning and policy research, with known gaps in smart meter coverage documented for users. The dataset can support diverse applications, including electricity load forecasting, socio-economic and behavioral energy research, distribution planning, and policy analysis in energy affordability and equity. Access is provided through an open repository, with accompanying documentation and scripts to facilitate reuse.

(RECON-SL)

This section describes RECON-SL’s linked household electricity records and longitudinal survey data, including its support for energy and policy research and documented smart-meter coverage gaps and limits to generalizability. It also covers consent, privacy protections, residual re-identification risk, and publicly available data-processing and exploratory-analysis code.

Survey Data

  • Across three waves, surveys track household characteristics and appliance stock, changes in usage and attitudes, and later affordability, tariff perceptions, payment practices, and energy security.The surveys were conducted over 15 months using tablet-based CAPI in Sinhala and Tamil, with efforts to maintain respondent consistency.Source: Survey Data · p. 2, Survey Data Collection · p. 4

Dataset Composition

  • The dataset covers 4,063 households, with monthly consumption for all and smart-meter records at 15-minute and 6-hour resolutions for 1,438 households.The complementary 6-hour files contain some features absent from the raw 15-minute stream and were used to address observed gaps; the dataset also includes 2,625 non-smart-metered households.Source: Dataset Composition · p. 2, Figure 3 · p. 3

RECON-SL’s Positioning Relative to Existing Datasets

  • RECON-SL combines multi-resolution utility records with three linked survey waves at thousand-household scale in a South Asian emerging-economy context, a combination uncommon among existing datasets.This combination enables load modeling across temporal scales and behavioral or policy analyses linking household characteristics and attitudes with consumption.Source: RECON-SL’s Positioning Relative to… · pp. 3–4

Suitable and Potential Use Cases

  • Potential uses include multi-scale load forecasting, behavioral analysis, distribution-level load profiling, solar-export research, and evaluation of affordability, equity, tariffs, and energy security.The dataset is publicly available through IEEE DataPort, Zenodo, and Kaggle and had more than 200 downloads at the time of writing; no peer-reviewed uses had been identified then.Source: Suitable and Potential Use Cases · p. 4, Prior uses of this dataset · p. 4

Study Population and Sampling

  • Stratified random sampling across LECO branches yielded 4,063 Wave 1, 3,500 Wave 2, and 3,397 Wave 3 participants, with overall retention above 80%.LECO serves about 500,000 households, primarily along Sri Lanka’s western coastal belt; the sample balances smart-metered and conventional-metered households.Source: BACKGROUND · p. 2, Study Population and Sampling · p. 4

Consistency and Internal Validation

  • Validation found smart-meter aggregates consistent with LECO billing records, 6-hour data in agreement with 15-minute readings, and consumption profiles within reasonable ranges.Survey retention reached 83.6%; collection quality checks included back-checks, audits, GPS verification, and cross-wave logic checks.Source: Consistency and Internal Validation · p. 5, Survey Data Quality · p. 5

Completeness of Meter Data

  • 15-minute records have 20.2% completeness against requested dates (35.5% within available dates), versus 47.7% and 51.7% for 6-hour records; 89.3% of 6-hour days cover at least half of households.Monthly billing data and all three survey waves have 100.0% completeness against requested dates; the reported smart-meter gaps reflect variable coverage.Source: Completeness of Meter Data · p. 4, Survey Wave 3 · p. 6, Table 2

Known Limitations

  • Results should not be generalized beyond LECO’s western coastal service areas; missing meter intervals limit fine-grained analysis, and household-level records cannot directly support appliance disaggregation or live operational forecasting.The dataset lacks precise household income, and self-reported expenditure may be subject to recall or estimation error.Source: Known Limitations · p. 6, Unsuitable and Cautionary Use Cases · p. 7

Ethics & Anonymization

  • Survey participation was consent-based, with withdrawal available; names and addresses were removed and household IDs pseudonymized, though linkage with external datasets may still enable re-identification.The dataset datasheet documents the ethical safeguards and anonymization procedures.Source: Ethics & Anonymization · p. 7

SOURCE CODE AND SCRIPTS

  • The LIRNEasia Lacuna repository publicly provides the code for cleaning, preprocessing, and initial analysis of the consumption and survey datasets.The repository also contains exploratory data analysis scripts.Source: SOURCE CODE AND SCRIPTS · p. 7

APPENDIX A: SURVEY METHODOLOGY NOTE · A.I. OVERVIEW · A.II. OVERALL APPROACH AND OBJECTIVES

The Survey Methodology Note supplements the dataset article with details on the household survey’s design and implementation. It covers sampling, fieldwork, quality control, and procedures supporting consistency across survey waves.

A.I. OVERVIEW

  • The note provides a detailed account of the household survey that complements the dataset described in the main article.Source: A.I. OVERVIEW · p. 9
  • It documents the survey’s sampling frame, fieldwork procedures, and quality-control protocols.Source: A.I. OVERVIEW · p. 9
  • It adds details on enumerator recruitment and training, questionnaire development, and piloting.Source: A.I. OVERVIEW · p. 9
  • It describes back-checking and monitoring procedures used during the survey.Source: A.I. OVERVIEW · p. 9
  • It addresses approaches to managing attrition across survey waves.Source: A.I. OVERVIEW · p. 9
  • The note is organized around overall approach and objectives, sampling and recruitment, instrument development, fieldwork administration, and quality control.Source: A.I. OVERVIEW · p. 9

A. Study Design and Implementation … B. Survey Respondents

The survey was designed to collect longitudinal household and energy-behavior data for linkage with electricity records, supporting analysis of demand-side measures relevant to affordability, efficiency, and policy in Sri Lanka.

A. Study Design and Implementation

  • Three face-to-face survey waves used Computer Assisted Personal Interviewing, and Wave 1 surveyed over 4,000 households across LECO’s seven branch areas.Wave 1 responses were later linked with electricity consumption records to form a combined dataset of household demand and behavioral drivers.Source: A. Study Design and Implementation · p. 9
  • Survey Research Lanka conducted fieldwork after competitive procurement, handling scripting, translation, pilot testing, training, and dataset delivery.LIRNEasia developed questionnaires, provided training, monitored fieldwork, and liaised with LECO for approvals and household contact information.Source: A. Study Design and Implementation · p. 9
  • A pause in Wave 1 fieldwork from January to July 2024 shortened the spacing between waves, so intervals were not uniform across households.Source: A. Study Design and Implementation · p. 9

B. Objectives

  • The survey aimed to link longitudinal data on household demographics, physical characteristics, appliance ownership, and energy behaviors with high-frequency smart meter data.The resulting dataset is intended to support analysis of behavioral nudges, tariff reforms, and other demand-side management solutions relevant to energy affordability, efficiency, and policy in Sri Lanka.Source: B. Objectives · p. 9

A. Survey Target Group

  • The target population comprised households served by LECO, located primarily along Sri Lanka’s south-western coastal belt.Source: A. Survey Target Group · p. 9

B. Survey Respondents

  • The household head was interviewed when available; otherwise, a suitable alternative able to provide household and electricity-use details was sought.Effort was made to interview the same household member in all three waves, but a suitable alternative was interviewed for Waves 2 or 3 when the original respondent was unavailable.Source: B. Survey Respondents · p. 9

C. Sample Frame & Procedure

The study drew households from LECO’s customer database using stratified random sampling by meter type and branch area. Survey participation declined across three waves, and data-retention failures left some surveyed smart-meter households without retained readings.

C. Sample Frame & Procedure

  • From LECO’s customer database, researchers drew 7,996 households through stratified random sampling by meter type and branch area to target 4,000 households split evenly by meter type.The target was 2,000 smart-metered and 2,000 non–smart-metered households.Source: C. Sample Frame & Procedure · p. 9
  • Wave 1 surveyed 4,063 households (1,788 smart-metered; 2,275 non–smart-metered), declining to 3,501 in Wave 2 and 3,398 in Wave 3 through attrition.The planned 2,000–2,000 split was not achieved because refusals exhausted the eligible smart-metered list, and replacements were not possible; retaining the predefined list was essential for consistent linkage to consumption records.Source: C. Sample Frame & Procedure · p. 9
  • Retained smart-meter readings were unavailable for 350 of the 1,788 surveyed smart-metered households, leaving 1,438 with readings in the final dataset.LECO system retention failures caused the missing consumption records, so those households were effectively treated as non-smart-metered for the final dataset.Source: C. Sample Frame & Procedure · p. 9

D. Selection of Sample LECO Customer Households … A.V. FIELDWORK ADMINISTRATION

The fieldwork used a stratified LECO household sample and three consented survey waves designed to track household changes. Questionnaire development, translation, scripting, and field pilots informed refinements to the research tools and field process.

D. Selection of Sample LECO Customer Households

  • LIRNEasia selected a predefined number of LECO customer households from each stratum and gave field teams identifying details for accurate household identification.The information provided included customer name, address, contact number, and meter type.Source: D. Selection of Sample LECO Customer… · p. 10
  • Wave 1 respondents were told the study involved three waves over about 12 months and asked to consent to participation and availability at the same household.Consent and availability were sought before interviewing began.Source: D. Selection of Sample LECO Customer… · p. 10

E. Household Replacement Procedures

  • If a household could not be reached, the team made at least three attempts before classifying it as noncontactable and replacing it within the same stratum.Attempts varied by time of day and day of the week, or were made by appointment.Source: E. Household Replacement Procedures · p. 10

A. Development, Translation & Scripting

  • Three separate questionnaires were developed to capture temporal changes, with Wave 1 covering household members, housing and rooms, appliances, electricity generation, cooking, and water heating.Wave 1 also collected demographics and staying habits, room characteristics, and appliance use.Source: A. Development, Translation & Scripting · p. 10
  • Waves 2 and 3 recorded changes from the previous wave; Wave 2 added appliance purchase details and clothing-care behavior, while Wave 3 covered device behavior, time-of-use awareness, and billing and payment habits.Follow-up questionnaires updated household membership, housing, rooms, appliances, and solar use.Source: A. Development, Translation & Scripting · p. 10
  • LIRNEasia developed and pilot-refined the English questionnaires, while SRL translated the final versions into Sinhala and Tamil and scripted them with logical and consistency checks.LIRNEasia checked translations; internal face-to-face piloting assessed length, flow, comprehension, cognitive difficulty, and question sensitivity.Source: A. Development, Translation & Scripting · pp. 10–11

B. Pilot Testing the Final Questionnaire

  • The pilot tested interview length, routing, phrasing, translations, question comprehension and sensitivity, and tablet functionality.Senior SRL professionals and LIRNEasia researchers led the field pilot.Source: B. Pilot Testing the Final Questionnaire · p. 11
  • Pilot observations led to improvements to both the field process and research tools.The final CAPI-scripted questionnaire was pilot-tested in an actual field setting among LECO households outside the survey sample.Source: B. Pilot Testing the Final Questionnaire · p. 11

A. Interviewer Training · B. Respondent Consent · C. Conducting Fieldwork

The survey used structured interviewer preparation and respondent consent procedures, while delays affected fieldwork timing and the spacing between waves.

A. Interviewer Training

  • Interviewers received one-day, questionnaire-focused training covering household definitions, recording timings, terminology, and the rationale for selected questions.Training included examples such as collecting the number of doors and windows.Source: A. Interviewer Training · p. 11
  • Training also covered field ethics, professional conduct, and handling customer issues; enumerators completed mock interviews before data collection.Interviewers were asked to make appointments and carry authentication documents during fieldwork.Source: A. Interviewer Training · p. 11

B. Respondent Consent

  • Before interviews, respondents were informed of the research objectives, that participation was voluntary, and that they could end the interview at any point.Source: B. Respondent Consent · p. 11
  • In wave 1, respondents were told they were joining a longitudinal study and would be interviewed in two further waves within the next year.Source: B. Respondent Consent · p. 12

C. Conducting Fieldwork

  • Wave 1 fieldwork began in November 2023, paused at the end of January 2024 after concerns from LECO, resumed in July, and finished by mid-October.Source: C. Conducting Fieldwork · p. 12
  • Because of delays and time limitations, wave 2 began in September 2024 overlapping wave 1, while wave 3 ran from late November 2024 into early January 2025.Source: C. Conducting Fieldwork · p. 12
  • Delays and tight timelines prevented consistent intervals between waves across households, but every household had at least a 3–4-week gap between waves.Source: C. Conducting Fieldwork · p. 12

D. Response Rate … Q3. Are you able to accurately tell me about the details related to the household?

The survey retained 83.6% of households from Wave 1 through Wave 3, with response and retention rates also reported for earlier waves and smart-meter segments. The supplied passages additionally describe survey quality checks and how the questionnaire defines the household head.

D. Response Rate

  • Wave 1 success was 40.6% overall, 44.7% among smart-meter households, and 37.9% among non-smart-meter households.The SRL team began with 7,996 contacts and later added 2,000 non-smart-meter household contacts due to a low hit rate.Source: D. Response Rate · p. 12
  • 86.2% of Wave 1 households (3,501) were re-contactable for Wave 2, an attrition of 13.8%.Table A3 summarizes Wave 2 contacts.Source: D. Response Rate · p. 12
  • 97.1% of Wave 2 households (3,398) were successfully re-contacted for Wave 3; retention was 96.4% for smart-meter and 97.6% for non-smart-meter households.At the end of Wave 3, the sample comprised 1,443 smart-metered and 1,955 non-smart-metered households.Source: D. Response Rate · p. 12
  • 83.6% of the 4,063 households contacted in Wave 1 were retained in Wave 3, with retention highest in Negombo and Galle and lowest in Nugegoda.Kelaniya and Kotte also had lower retention than Negombo and Galle.Source: D. Response Rate · p. 12

A.VI. QUALITY CONTROL (QC) MEASURES

  • Survey quality checks included supervisor accompaniment, interview back-checks, spot and voice checks, GPS verification, duration and outlier checks, CAPI filters, and interviewer debriefings.Back-checks covered 18% of each interviewer’s interviews, spot checks covered 10% of the sample, and telephonic back-checks covered 80% of interviews.Source: A.VI. QUALITY CONTROL (QC) MEASURES · pp. 12–13

Q1. Is this premise used as a household or for business/commercial purposes?

  • The questionnaire defines the household head as the person making the most important household financial decisions or contributing the most to day-to-day expenses.The supplied passage labels this item Q2.Source: Q1. Is this premise used as a… · p. 14

Q4. Do you consent to participate in this study and to provide your data to LIRNEasia? … Q4. Are you able to accurately tell me about the details related to the household?

This questionnaire section documents household and dwelling characteristics, including household-member demographics, room and construction details, and socioeconomic and meter context. It also covers electricity and appliance use, household energy sources, and water and cooking practices.

Q6. Can you please tell me how many electricity meters are in your house?

  • The survey asks about meter count, home ownership or renting, renter/boarder occupancy, and respondents’ awareness of boarders’ electricity use.The questions distinguish boarders sharing the meter and ask whether respondents know details such as their appliances and usage.Source: Q6. Can you please tell me how many… · p. 14

Q10. Which of the following time periods best describes when your house was built?

Q13. How many storeys does your house have?

Q17. Relationship to the household head

Q25. Highest level of education

  • The survey records current educational attendance, highest education attained, and each member’s main activity.Education options range from school grades to G.C.E. qualifications and degrees; activity options include employment, seeking work, retirement, household activities, and study.Source: Q23. Marital status · p. 15, Q25. Highest level of education · p. 15

Q27. Main occupation

Q35. What type of business activity is being carried out in this household?

Q41. What is the main material used to build the outside walls of your house?

Q44. For what purposes do you mainly use this section of the household?

Q54. How many light bulbs do you have in this section of the household?

Q64. What is the type of air conditioning used in the room?

Q74. Does your household have a backup generator to generate electricity if needed?

Q76. Is your solar system on-grid or off-grid?

Q83. How does your household receive water?

Q92. What is the Occupation of the Chief Wage Earner in your house?

Q5. Do you consent to participate in this study and to provide your data to LIRNEasia? … Q3. Are you able to accurately tell me about the details related to the household?

This survey section documents household composition and housing conditions, including changes over time, and records related electricity use, appliance and energy practices, expenditures, meter type, and respondent continuity across waves.

Q8. Please tell me how many electricity meters are there in your house now?

  • The questionnaire distinguishes owner-occupied, rented, and rent-free housing and identifies whether renters or boarders share the household’s electricity meter.It also asks whether boarders were present at the first interview or arrived later.Source: Q8. Please tell me how many… · pp. 20–21

Q12. Since when did the new renters/boarder(s) start living in this household?

  • The survey asks when new renters or boarders arrived, how much respondents know about their electricity use, and whether the household’s physical structure changed since the first interview.Structural changes include additions, completed construction, new floors, or dismantled parts.Source: Q12. Since when did the new… · p. 21

Q16. Which of these best describes your house now?

Q24. Relationship to the household head

Q32. Highest level of education

Q36. Employment status of the main occupation

Q42. What type of business activity is being carried out in this household?

Q51. For what purposes do you mainly use this section of the household?

Q54. What is the main material used for the roof of this section of the household?

Q66. What is the wattage (W) of the light?

Q82. Do you have a solar panel to generate electricity at your home?

Q111. Size of TV

Q117. Which of the following best describes your behaviour of ironing clothes?

B.III. Survey Wave 3

  • Wave 3 asks whether the respondent is the same person as in Wave 1 and/or 2 or a different household member.Source: B.III. Survey Wave 3 · p. 28

Q4. Do you consent to participate in this study and to provide your data to LIRNEasia? … Q113. Which of the following can most accurately describe your knowledge in electricity bill calculation?

This survey section combines information on household composition, demographics, housing, and renters’ electricity arrangements with measures of appliance and solar-system ownership and use, conservation behavior, and metering. It also covers time-of-use billing preferences, electricity targets, bill-calculation knowledge, and payment practices.

Q7. Please tell me how many electricity meters are there in your house now?

  • The survey asks about home ownership or renting, boarders sharing the meter, when renters arrived, and how much respondents know about boarders’ electricity use.It distinguishes renters who share living conditions from boarders in rooms attached to the household and asks whether renters were present at the second interview.Source: Q7. Please tell me how many… · p. 28, Q11. Since when did the new… · p. 29

Q15. Which of these best describes your house now?

Q16. How many storeys does your house have?

Q25. Ethnicity

Q29. Highest level of education

Q31. Main occupation

Q35. During the last week, did this member go out of the home for employment purposes?

Q41. Employment status of main occupation.

  • For renters or boarders sharing the meter, the survey asks whether electricity is charged at a fixed monthly amount, according to the bill, or as part of a combined utilities charge.Source: Q41. Employment status of main… · p. 31
  • For rented households, it asks how electricity payments are handled and whether business activity uses electricity from the household’s meter.Payment options include full or fixed payments, variable payments linked to bills, combined utility charges, or no separate electricity payment.Source: Q41. Employment status of main… · p. 31

Q49. Duration of usage in hours during the last week

  • The questionnaire inventories 77 appliance types used in working condition at least once, including appliances used by boarders or renters.The list spans cooking, refrigeration, entertainment, information and communications, security, vehicles, water systems, and other electrical equipment.Source: Q49. Duration of usage in hours during… · p. 32

Q53. Are there any changes to the number of ____ (room type)?

Q66. And how many of these bulbs did you use in the last week?

Q89. Do you have a solar panel to generate electricity at your home?

Q99. Which appliances/devices have you and your household members reduced or stopped using consciously?

Q105. Why have you not used “Time of use metering”?

  • The survey explained peak, daytime, and off-peak time-of-use rates, then asked whether respondents would use them and shift appliance use off-peak for at least 20% bill savings.It also described a fixed monthly charge alongside the time-period charges.Source: Q105. Why have you not used “Time of… · p. 36

Q109. What is your target type?

Q113. Which of the following can most accurately describe your knowledge in electricity bill calculation?

APPENDIX C: DATASHEET

The datasheet explains that RECON-SL was created to fill a Sri Lankan gap in granular household electricity data and documents its development, collaboration, and funding.

Purpose

  • The dataset addresses a Sri Lankan gap in granular electricity data for AI-based demand-side management, integrating high-resolution consumption with household demographic, social, and physical context.At the time, no comparable dataset existed in Sri Lanka, and few datasets globally offered the same integration.Source: Appendix C · p. 37

Creators and collaboration

  • LIRNEasia’s Data, Algorithms, and Policy Team developed the dataset in collaboration with LECO, which provided de-identified consumption records and supported the longitudinal survey.LIRNEasia is a Sri Lanka-based public policy think tank.Source: Appendix C · p. 37

Funding

  • Lacuna Fund funded the dataset under the grant “20-Month Dataset on Household Electricity Consumption and its Drivers, Collected via Meter Readings, and Longitudinal Survey” (Grantee #55 (19497.59)).Lacuna Fund supports labeled datasets addressing urgent problems in low- and middle-income contexts.Source: Appendix C · p. 37

How many instances are there in total (of each type, if appropriate)? · Any other comments?

RECON-SL is a weighted sample of 4,063 LECO households, combining electricity readings with longitudinal survey data; it represents LECO customers, not all Sri Lankan households. The documentation describes collection and cleaning procedures, missing-data and noise limitations, and consent and anonymization safeguards.

How many instances are there in total (of each type, if appropriate)?

  • 4,063 households—1,438 smart-metered and 2,625 non-smart-metered—were sampled from LECO’s approximately 500,000 households and weighted for geographic representation, but do not represent customers of other providers.The dataset describes the sample as representative of LECO customers, not Sri Lanka’s entire population.Source: How many instances are there in total… · pp. 37, 39
  • Electricity data include 15-minute and 6-hour smart-meter readings, monthly readings, and voltage, current, and frequency; surveys provide household information, with no single predefined target label.Potential targets include consumption, meter measurements, demographics, appliance use, and energy-security responses.Source: How many instances are there in total… · pp. 37–38
  • Smart-meter time-series gaps mainly result from transmission failures, while some survey data are missing because of incomplete responses and data-entry errors.Potential noise also includes recall bias, respondent misunderstandings, sensor or transmission failures, storage limits, and anomalous readings; preprocessing was applied.Source: How many instances are there in total… · p. 38

Any other comments?

  • LECO supplied observed meter data, while household surveys collected self-reported demographics and behavioral information from 15 November 2023 to 12 January 2025; consumption records span 2023-01-01 to 2024-12-23.Survey responses were checked through back-checks, supervisor accompaniments, and pilot testing; meter data were preprocessed for missing values, anomalies, and inconsistencies.Source: Any other comments? · pp. 39–40
  • Households were selected from the LECO customer database using stratified random sampling by meter type and then LECO branch area.Source: Any other comments? · p. 39
  • Participants received study and data-linkage explanations and gave consent; they could withdraw at any time or for specific uses without consequences.Participants were also told interviews might be audio-recorded with permission and were given contact routes to request data removal.Source: Any other comments? · p. 40
  • The data were anonymized to prevent direct or indirect identification, although household and demographic attributes may be sensitive; a comprehensive data protection assessment had not yet been conducted.Source: How many instances are there in total… · p. 39, Any other comments? · pp. 40–41
  • Survey data were cleaned through response validation, format standardization, appropriate imputation of missing responses and meter gaps, longitudinal alignment, and de-identification.Source: Any other comments? · p. 41

6) Electricity Consumption Data · Any other comments?

The dataset was cleaned and aligned, but raw data were not retained. It is publicly available across several platforms, with usage tracking and research applications still developing.

6) Electricity Consumption Data

Any other comments?

  • No repository currently links known dataset uses, but those uses will be documented in the project GitHub repository.Source: Any other comments? · p. 41

What other tasks could the dataset be used for? · Any other comments?

The dataset supports a range of analytical and predictive energy tasks, but its geographic scope, household-level metering, proxy income measures, and historical data with transmission gaps constrain some uses. It is publicly available under CC BY 4.0 through IEEE DataPort, Zenodo, and Kaggle.

What other tasks could the dataset be used for?

Any other comments?

  • Third parties may access, use, share, and modify the dataset under CC BY 4.0, provided they give appropriate credit.Source: Any other comments? · p. 42
  • The dataset is hosted on IEEE DataPort, Zenodo under the same DOI, and Kaggle; its DOI is 10.21227/n1dk-q860.Source: Any other comments? · p. 42

When will the dataset be distributed? · Any other comments? · Who will be supporting/hosting/ maintaining the dataset?

The dataset is already publicly available under CC BY 4.0, with no imposed IP-based restrictions; hosting is supported on IEEE DataPort, Zenodo, and Kaggle.

When will the dataset be distributed?

Who will be supporting/hosting/ maintaining the dataset?

How can the owner/curator/manager of the dataset be contacted (e.g., email address)? · Any other comments?

The dataset will not gain new instances or new versions, though corrections or removals may be made and documented on GitHub. Household information is de-identified, with no retention restrictions; extension is possible only through a non-straightforward collaborative process.

How can the owner/curator/manager of the dataset be contacted (e.g., email address)?

Loading 2609.28783v1…