A Systematic Literature Review on Machine Learning Techniques for Predicting Household Water Consumption

Zenodo Repository Metadata and Description

Title: Supplemental Material for "A Systematic Literature Review on Machine Learning Techniques for Predicting Household Water Consumption"

Description

This repository contains the dataset and methodological artifacts developed for a Systematic Literature Review (SLR) investigating Machine Learning (ML) applications in residential water demand forecasting. The review synthesizes evidence from 80 primary studies published between 2009 and 2024, following the PRISMA-2020 guidelines.

Repository Contents

DataExtraction.csv: The master evidence matrix containing metadata, objectives, and findings for all 80 primary studies.

Taxonomy Maps: Standardized mapping files that document how raw textual terms from the literature were normalized into canonical categories.

Supplementary Material: The file provides detailed methodological documentation and extended analyses that support the results reported in the main manuscript, while preserving conciseness in the primary text.

The supplementary material includes:

A detailed methodological workflow and execution timeline, describing the implementation of the seven-step SLR protocol in accordance with PRISMA guidelines, including database search, screening, quality assessment, data consolidation, and analysis.

A complete list of the 80 primary studies analyzed in the review, including publication type and reference identifiers.

An extended characterization of machine learning model properties (RQ1), documenting the conceptual taxonomy used to classify model capabilities and methodological features across studies.

A qualitative synthesis of reported benefits associated with evaluation metrics (RQ3), complementing the quantitative results presented in the main article.

A citation-based contextual analysis, including total citations and citations per year (CPY), stratified by model families, to identify influential studies and thematic trends within the literature.

All quantitative mappings, cross-references, and classifications described in this document are aligned with the evidence matrix made publicly available alongside this repository. This supplementary file is intended to enhance transparency, reproducibility, and methodological rigor for researchers interested in machine-learning-based residential water demand forecasting.

Key Research Insights

Methodological Evolution: A capability shift around 2020 where Deep Learning models (notably LSTM) displaced traditional methods.

Predictor Patterns: Climatic and historical consumption dominate, with limited use of socio-economic variables due to privacy constraints.

Evaluation Standards: RMSE and MAE are the most widely used evaluation metrics.

Practical Tools for Researchers

Model-Predictor Crosswalk: A co-occurrence matrix of model families and predictor categories.

Prescriptive Decision Tree: A heuristic framework guiding model selection based on data and computational constraints.

# Zenodo仓储元数据与附属说明

## 标题
《“家庭用水量预测机器学习技术系统综述”补充材料》

## 说明
本仓储包含为一项针对住宅需水预测领域机器学习（Machine Learning, ML）应用的系统综述（Systematic Literature Review, SLR）所开发的数据集与方法学成果。本综述遵循PRISMA-2020指南，对2009年至2024年间发表的80项原始研究进行了证据整合。

## 仓储内容

### DataExtraction.csv
核心证据矩阵文件，涵盖全部80项原始研究的元数据、研究目标与研究结果。

### 分类映射表
标准化映射文件，用于记录文献中原始文本术语如何被归一化至标准分类体系。

### 补充材料
本文件提供了详细的方法学说明与拓展性分析内容，用以支撑主稿件中报告的研究结果，同时保证主文稿的简洁性。补充材料包含以下内容：
1. 详细的方法学工作流与执行时间表：详述了遵循PRISMA指南制定的七步SLR方案的实施过程，涵盖数据库检索、文献筛选、质量评估、数据整合与分析等环节。
2. 本次综述所分析的80项原始研究完整列表，包含文献类型与参考文献标识。
3. 针对机器学习模型属性的拓展性特征描述（研究问题1/RQ1）：记录了用于分类各研究中模型能力与方法学特征的概念分类体系。
4. 针对评估指标相关报告收益的定性整合（研究问题3/RQ3）：作为主文章中定量结果的补充内容。
5. 基于引用的语境分析，包含总被引频次与年均被引频次（Citations Per Year, CPY），按模型家族分层，用以识别领域内高影响力研究与主题趋势。

本文档中描述的所有定量映射、交叉引用与分类结果，均与本仓储同步公开的证据矩阵保持一致。本补充文件旨在为关注基于机器学习的住宅需水预测领域的研究者提升研究透明度、可复现性与方法学严谨性。

## 核心研究结论
1. 方法学演进：2020年前后出现能力范式转变，深度学习模型（Deep Learning，尤以长短期记忆网络（Long Short-Term Memory, LSTM）为代表）取代了传统机器学习方法。
2. 预测因子模式：气候因子与历史用水量数据为最常用的预测变量，由于隐私限制，社会经济变量的应用较为有限。
3. 评估标准：均方根误差（Root Mean Squared Error, RMSE）与平均绝对误差（Mean Absolute Error, MAE）为应用最广泛的评估指标。

## 面向研究者的实用工具
1. 模型-预测因子对应表：模型家族与预测因子类别的共现矩阵。
2. 指导性决策树：基于数据与计算约束指导模型选择的启发式框架。