Analyzing Patterns and Predictive Models of Energy and Water Consumption in Schools

Abstract

Schools are major consumers of energy and water, significantly influencing environmental sustainability and operational budgets. This study presents a comprehensive review of global trends in energy and water consumption in school buildings, identifying key factors that shape usage patterns, such as the geographic location, climate, building characteristics, and occupancy levels. A particular focus is placed on the role of predictive models in enhancing resource efficiency. The review found that energy consumption in schools varies widely, with heating, lighting, and cooling systems being the primary contributors. In contrast, research on water consumption—especially predictive modeling—is notably scarce, with no studies found that focused specifically on school buildings. This highlights a critical gap in the literature. This study evaluated the existing predictive approaches, including regression analyses, machine learning algorithms, and statistical models, which offer valuable tools for forecasting consumption and guiding targeted efficiency interventions. The findings underscore the urgent need for data-driven strategies to support sustainable resource management in educational facilities.

1. Introduction

Buildings have been identified as responsible for over 40% of energy usage in many nations globally [1,2,3]. Public buildings, particularly schools, provide an excellent potential for implementing energy-efficient measures. This is due to their high share of the overall building stock, resulting in a significant contribution to both total energy consumption and national budgets [4]. Additionally, schools bear a substantial social responsibility in the building sector due to their educational role [5]. The literature also shows that a school’s energy costs rank as the second largest cost for the school, behind the wages of teachers and employees [6]. Energy consumers at schools exhibit distinct characteristics compared to the users of public and residential buildings, particularly in terms of their activities and the duration of occupancy within the building itself. Also, employees and students lack the same financial incentive to conserve energy at the school as they have in their own homes, since they possess limited knowledge about the quantity and cost of the energy utilized, and rooms and devices are mostly shared [7]. Given the substantial and fluctuating water and energy usage in schools, it is becoming more imperative to implement techniques aimed at minimizing the consumption of these resources [8].

The emphasis on educational buildings as a significant portion of non-residential buildings underscores the necessity of targeting energy efficiency efforts at these facilities [9,10,11]. The fundamental goal for building energy management and facility managers is to forecast and predict energy usage in buildings [12].

Several countries have acknowledged substantial consumption in the area of non-residential buildings. For example, in the United Kingdom (UK), there are 2 million buildings in the non-residential sector, which contribute to 19% of the country’s total CO₂ emissions and present a substantial potential for reducing emissions [13]. In Canada, school buildings account for 30% of the energy consumed by the public sector [14]. Furthermore, school buildings account for around 13% of the overall energy usage in the USA, 4% in Spain, and 10% in the UK [15,16].

In recent decades, numerous techniques have been proposed for forecasting energy use in building construction. The majority of case studies employ historical energy usage data to construct the prediction models. The techniques devised for forecasting building energy consumption can be categorized into two distinct groups: statistical methods and artificial intelligence [17]. Olu-Ajayi et al. conducted a comprehensive analysis of data-driven methods for predicting building energy consumption. Their evaluation emphasized that artificial neural networks (ANNs) consistently outperformed statistical tools like multiple linear regression (MLR) in the majority of the trials. However, MLR demonstrated excellent results in some situations, such as forecasting the annual energy consumption [18].

Several diverse factors impact the energy usage of a school building [6,15,19]. Therefore, when developing an energy usage forecasting model, it is crucial to take these elements into account. Grid management needs to forecast energy consumption in built facilities in order to save electricity, ensure effective use, and cut down on waste [20,21]. Nonetheless, generating reliable predictions is difficult due to unpredictable circumstances and the inherent disorder of data, and the methods utilized often yield misleading projections [22].

The energy usage intensity (EUI) is an important factor in energy performance analyses, as it measures the energy consumption per unit of useable area (kWh/m²/year), regardless of the building type [9]. Nevertheless, certain studies have also examined the measurement of EUI for school buildings by quantifying the energy usage per student (kWh/student/year).

Upon conducting a thorough analysis of the relevant literature, the authors discovered that just one review paper by Dias Pereira et al. addressing energy usage in schools was identified, published in 2014 [6].

The objective of this study was to perform a thorough review of energy and water usage trends in school buildings. By reviewing the existing literature and previous research, this study aimed to identify various consumption patterns depending on the location, school type, unit of measure, and sample size and to examine predictive models for energy consumption. Ultimately, this study sought to provide insights into the sustainability, efficiency, and resilience of school infrastructure, contributing to the development of more environmentally friendly and cost-effective education.

The remainder of this paper is structured as follows: Section 2 presents the methodology of the paper, Section 3 provides relevant research on the consumption of heating and electrical energy in schools, Section 4 provides relevant research on water consumption in schools, Section 5 presents the developed models for predicting heating and electrical energy consumption in schools, Section 6 presents the developed models for predicting water consumption in educational buildings, Section 7 provides a discussion, and in Section 8, there is a conclusion.

2. Methodology

The methodology for reviewing and analyzing patterns and predictive models of energy and water consumption in schools and educational buildings followed a systematic and structured approach. Relevant studies were identified, filtered, and extracted from reputable academic databases, including Web of Science and Scopus. The review focused on peer-reviewed scientific papers published in English that examined energy and/or water consumption in school settings, with particular attention to predictive modeling techniques. The search included studies published between 1999 and 2024, with no formal time restriction applied, although more recent studies were prioritized to reflect current technologies and practices. Only one conference paper was included; practical reports, non-peer-reviewed sources, and non-English publications were excluded to ensure academic rigor. A keyword-based search strategy was applied using terms such as “school buildings,” “energy consumption,” “water consumption,” “predictive modeling,” “regression,” and “machine learning.” Studies were screened in stages, first by title and abstract, and then by a full-text analysis to assess their relevance and methodological quality. The inclusion criteria required the studies to provide measurable data on resource consumption and/or describe a predictive model applicable to school buildings. The studies were further categorized based on the building type (e.g., primary, secondary, preschool), geographic region, consumption type (energy or water), and modeling approach. This classification formed the basis for the comparative analysis and synthesis presented in this review.

3. Heating and Electrical Energy Consumption in Schools

Butala and Novak performed a thorough examination of energy and interior environmental audits of 24 school buildings in Slovenia in 1999. The audits uncovered that these facilities demonstrate increased energy use and experience insufficient indoor air quality, as reported by 60% of the evaluated students. A recent discovery indicated that the boilers and heat exchangers used in district heating systems have an over-capacity of 57% in relation to their nominal heating output. The average yearly energy consumption for heating, sanitary hot water, and lighting was determined to be 192 kWh/m²/year. Furthermore, the school buildings experienced heat losses that were beyond the acceptable standards by 89% [23]. Kim et al. conducted a study to ascertain school buildings’ optimal quantitative energy consumption level for maintaining a comfortable school environment and achieving good performance while ensuring effective energy utilization and control. An analysis was conducted on the energy consumption of 10 elementary schools in Daegu, a city in the southern region of South Korea. The analysis considered the data from January 2006 to December 2010 and examined the energy consumption in terms of the year, unit area, and per capita. This analysis aimed to provide specific values for the energy-saving objectives of elementary schools in South Korea. The energy consumption per unit study area of the elementary schools in South Korea in 2010 were found to be 1040 MJ/m²/year for electricity, 92 MJ/m²/year for oil, and 325 MJ/m²/year for gas. Since one megajoule equals 0.278 kWh, the consumption values were the following: 289 kWh/m²/year for electricity, 26 kWh/m²/year for oil, and 90 kWh/m²/year for gas [24].

Wang conducted a study on the final energy usage in 67 senior high schools, 62 junior high schools, and 102 elementary schools in Taiwan. Their energy use intensity values (kWh/m²/year) were 55.8, 22.5, and 20.1. Their energy usage per person (kWh/person/year) was 1163, 469, and 465, respectively. It was found that senior high schools utilized a significantly higher amount of energy compared to elementary and junior due to their larger size, air-conditioned classrooms, and additional facilities. Also, private schools demonstrated a much higher energy consumption, potentially due to their superior learning settings, teaching equipment, and larger average class sizes. Three MLR models were developed to estimate the total energy consumption (kWh/year), energy usage intensity, and energy use per person in school buildings. This study showed that air conditioning and lighting significantly impact the amount of electricity used in school buildings. Furthermore, it offers various practical energy-saving methods for administrators to assess their school’s energy usage and improve the energy efficiency. The authors propose that the findings of this study could be used as a guide by government authorities in developing energy conservation policies for school buildings [25].

Antunes and Ghisi assessed the potential for water and energy saving in public schools located in the southern area of Brazil. Water usage data were gathered from 62 schools, while energy consumption data were gathered from 100 schools. The energy consumption data collected spanned from January 2016 to May 2017. In high schools, the average energy consumption was 7.15 kWh/student/month, while middle schools had an average of 5.30 kWh/student/month. These data were gathered in conjunction with the student count, hours worked, and school type. Significant disparities were seen, ranging from 0.31 to 66.47 kWh/student/month [8]. Chung and Yeung conducted a survey on the energy use in 121 secondary schools, representing 25.6% of all secondary schools in Hong Kong. Their study revealed that the mean energy usage per school was 529,925 kWh and 105,610 kWh/m²/year [26]. Katafygiotou and Serghides conducted an in-depth study of the average energy usage of schools in Cyprus by utilizing questionnaires, field inspections, and interviews with school managers and technical staff. The actual energy usage of the school buildings was assessed by comparing it with their construction parameters using monthly bills, and the study determined that schools have an average yearly consumption of 62.75 kWh/m² [15]. Jurišević et al. analyzed the specific thermal energy consumption of educational buildings in Kragujevac, Serbia. The buildings were categorized by educational level, including preschool buildings, primary and secondary schools, and university buildings (faculties). Data on thermal energy consumption were collected over several heating seasons in order to minimize the impact of seasonal climatic variations. It was found that kindergartens and primary schools should be prioritized for energy renovation, as their average specific heat consumption is the highest (186 kWh/m²/year and 176 kWh/m²/year, respectively) [27]. Daly et al. conducted an analysis of the energy consumption statistics for a total of 3701 public primary schools, and the average energy consumption was found to be 38.0 kWh/m²/year and 542 kWh/student/year [28]. Corgnati et al. undertook a field survey to collect, analyze, and assess data on the energy usage for heating in a sample of 138 buildings (117 high schools, 9 office buildings, and 12 houses for school caretakers) in the Provincia di Torino. The findings showed that the adjusted conventional heat supply and the annual measured heat supply have a satisfactory correlation. However, there is a notable discrepancy between the predicted and real energy usage, as seen by the monthly heat supply statistics. Variability in the monitoring intervals, which may result in inaccurate meter readings at the end of each month, is the cause of the observed result. The authors claim that this methodology works effectively for assessing large building stocks over an extended period of time. It is useful for calculating the annual and overall cost of energy services [29].

Beusker et al. conducted an empirical study that examined the factors that affect the amount of thermal energy used in municipal schools and sports facilities. Their investigation was conducted using a randomly selected sample of 105 properties in Stuttgart. Their study revealed that the energy usage in the examined schools varied between 31 and 205 kWh/m²/year, with an average consumption of 93 kWh/m²/year [30]. Raatikainen et al. conducted a study to evaluate and compare the electrical energy and heating consumption of six educational facilities in Kuopio, situated in Eastern Finland. The selected schools were built at different times, and their ventilation and building automation systems also lacked consistency. Data regarding the hourly consumption of energy were acquired from the local energy provider. According to the analysis, newer school buildings had a superior energy efficiency compared to older ones. The main novelty of their study was the utilization of hourly smart metering usage data for both electricity and district heating. The utilization of sophisticated computer methods facilitated the analysis of complex multivariate data, leading to an enhanced comprehension of the buildings’ consumption patterns and energy efficiency [31].

Thewes et al. conducted field research to investigate the energy use of 68 school buildings in Luxembourg. A comprehensive analysis of the electricity and heat energy use was carried out, allowing for a detailed examination of specific energy properties. The study revealed that new schools in Luxembourg exhibited a higher use of primary energy compared to older school buildings in other European nations [32]. In 2012, Kim et al. conducted a study on the energy use of nine schools in Korea. The study revealed that the energy usage of each school ranged from 400 to 1750 MWh/year, with variations based on the size of the school. Approximately 82% of the total energy usage at the assessed schools was attributed to electric power use, while liquefied natural gas and kerosene accounted for 14% and 4%, respectively. The schools that utilized fans for cooling exhibited a markedly reduced energy consumption in comparison to the schools that relied on an electric heat pump. The average annual energy consumption per unit area of the schools ranged from 67 to 240 kWh/m²/year, with significant variations seen between different schools. The studied schools had an average annual energy consumption of 133 kWh/m²/year [24]. The authors also continued their research in 2019 [33]. Hernandez et al. analyzed a sample of Irish primary schools where they used questionnaires to obtain information regarding the energy consumption of the schools and received 88 responses, which were used for a detailed analysis. It was found that the average annual thermal energy consumption was 96 kWh/m²/year [34].

Santamouris et al. performed a detailed energy survey on 10 schools’ energy efficiency and performance and their global environmental quality. The authors found that the mean annual energy consumption for heating was 57 kWh/m²/year and 20 kWh/m²/year for electricity [35]. Hong et al. analyzed the energy consumption in 7731 primary and secondary schools in England, where the results showed that the average consumption for primary schools was 166 kWh/m²/year, and for secondary schools, it was 172 kWh/m²/year [14,36]. Attia et al. created a dataset on the energy performance and two benchmark models for simulating the performance of high-performing schools in Belgium. Their study reported an average EUI of 59 kWh/m²/year for primary schools and 42 kWh/m²/year for secondary [37].

Table 1. Summary of the previous research regarding heating and electrical energy consumption in schools.

Reference	Year	Country	Type of School	Unit	Consumption Amount	Type of Energy	Sample Size
Butala and Novak [23]	1999	Slovenia	Secondary	kWh/m2/year	192	Heating, hot sanitary water, and lighting	24
Santamouris et al. [35]	2007	Greece	Not specified	kWh/m2/year	57	Heating	10
					20	Electricity
Hernandez et al. [34]	2008	Ireland	Primary	kWh/m2/year	96	Heating	88
Beusker et al. [30]	2012	Germany	Elementary and secondary with sport facility	kWh/m2/year	93	Heating	105
Hong et al. [36]	2013	England	Primary	kWh/m2/year	166	Heating	7731
			Secondary		172
Kim et al. [24]	2012	South Korea	Primary	kWh/m2/year	289	Electricity	10
					26	Oil
					90	Gas
Katafygiotou and Serghides [15]	2014	Cyprus	Secondary	kWh/m2/year	62.75	Heating and electricity	156
Jurišević et al. [27]	2018	Serbia	Primary	kWh/m2/year	176	Heating	14
			Secondary		136.37		8
Antunes and Ghisi [8]	2019	Brazil	High	kWh/student/month	7.15	Electricity	100
			Middle	kWh/student/month	5.30
Kim et al. [33]	2019	South Korea	Middle	kWh/m2/year	133	Heating and electricity	9
Wang [25]	2019	Taiwan	Senior high	kWh/m2/year	55.8	Liquefied petroleum gas, natural gas, heavy fuel, light fuel, and electricity	67
			Junior high	kWh/m2/year	22.5		62
			Elementary	kWh/m2/year	20.1		102
Chung and Yeung [26]	2020	Hong Kong	Secondary	kWh/m2/year	105.61	Electricity	121
Attia et al. [37]	2020	Belgium	Primary	kWh/m2/year	59	Total	30
			Secondary		42
Daly et al. [28]	2022	Australia	Primary	kWh/m2/year	38	Gas and electricity	3701
				kWh/student/year	392

provides a summary of the analyzed research.

4. Water Consumption in Schools

Since educational buildings comprise a sizable fraction of non-residential buildings, it is crucial to focus water-saving initiatives on these facilities [9,10,11]. Also, the availability and usage of water in schools significantly impact the health and cleanliness of the learning environment for children [38]. An excessive quantity of water in a school can be attributed to improper utilization or wastage. Nevertheless, minimal water consumption may not align with the concepts of health and sanitation [39]. Despite its importance, research on school water consumption is relatively rare worldwide, as Morote et al. noted in their work [40].

Cheng and Hong conducted a survey on water consumption in primary schools with the aim of promoting water conservation. At first, a database was established to record the water consumption in elementary schools. The questionnaires were disseminated to individual primary schools through postal delivery. Each response was examined for the school’s geographic location, the number of pupils and teachers, the male-to-female ratio, and other data. The analyzed data were gathered from 112 primary schools located in Taipei City, Taiwan. The appropriateness of the assessment technique for water use has been confirmed, encompassing the WCLoad and WCIndex categories. The study revealed that the overall average water use per person per year in the schools under analysis was 15.27 m³/person/year. The authors emphasize the tool’s value and its potential for enhancing water conservation in primary schools [39].

Schultt et al. conducted a study on 26 public elementary, middle, and high schools in southern Brazil to determine the factors that affect water use in schools. The data collection methods were in situ measurements and questionnaires, which the principals answered. The authors found that the average monthly water consumption was 111.45 m³/school/month, while the average water consumption per student was 8.833 L/student/day. The monthly water use of schools was predicted using a multiple linear regression analysis. The variables were selected using a stepwise process and a correlation analysis. Ultimately, it was demonstrated that the quantity of students and the utilization of restrooms were crucial determinants of the monthly water consumption [41].

Farina et al. integrated quantitative data on water consumption. Six hundred buildings in Bologna, Italy, were monitored for over five years. Their study examined the daily consumption rate of three types of schools—nurseries, kindergartens, and elementary schools—and investigated the relationship between consumption and building occupants. The authors concluded that the difference in school types was directly related to the difference in the purpose of the school and the age of the students. They found that elementary school pupils resembled office building occupants in terms of their presence during operating hours and their primary usage of water in restrooms and for general purposes. Ultimately, the reasonable fundamental requirement for water was calculated to be 48 L per preschool student per day and 18 L per elementary school student per day [42].

Almeida et al. conducted a study that specifically examined 23 school buildings on Portugal’s northern coast. All the buildings in the sample were in their original state and primarily built recently; over 40% of the sample was less than 20 years old [43]. Each facility’s energy and water costs were acquired from the corresponding monthly invoices supplied by the school board over one year. They discovered that electricity is the primary operational expense, with water being the second most significant. Additionally, it was observed that the electricity usage exhibited a noticeable seasonal fluctuation, with an increased demand during the winter months. However, no discernible trend was seen for water consumption [44].

Morote et al. studied water use at elementary and secondary schools in Alicante, Spain, from 2000 to 2017. The researchers acquired the outcomes using surveys that were distributed to school principals. Out of the 88 requests, only 14 principals consented to participate in the study. The study’s authors emphasized that the most significant restriction was the low response rate, even though they contacted the principals three times. The questionnaire consisted of five sections: school attributes, water usage and administration, garden attributes, views of water consumption, and initiatives to conserve water. Additionally, data were collected from a water company regarding the schools that participated in the survey, including the cost of water measured in euros per cubic meter (EUR/m³). It was found that water usage within the school was influenced by several aspects, including the student and employee population, the watering of outdoor spaces, the type of gardens, the efficiency of watering systems, the use of water-saving equipment, and cleaning activities [40].

Nunes et al. sought to identify the criteria for rational water utilization in public schools in Brazil. The technique comprised the following components: choosing a school, identifying all areas where water is being used in the school, identifying any instances of water leakage or malfunction, administering surveys to evaluate consumers’ perception, and calculating the consumption indicator, leakage index, and user perception index to promote efficient water usage. The authors highlighted that, although the water use may not have been as substantial as previously reported in the literature, the results indicated that the schools nevertheless encountered notable physical water losses and had a limited understanding of water efficiency [45]. Previous research regarding water consumption in schools is provided in

Table 2. Summary of the previous research regarding water consumption in schools.

Reference	Year	Country	Type of School	Unit	Consumption Amount	Sample Size
Cheng and Hong [39]	2004	Taiwan	Primary	m3/person/year	15.27	112
				L/student/day	30
Farina et al. [42]	2011	Italy	Preschool	L/student/day	48	600
			Elementary		18
Nunes et al. [45]	2018	Brazil	Not specified	L/student/day	8.58	6
Schultt et al. [41]	2019	Brazil	Elementary, middle, and high	m3/school/month	111.5	26
				L/student/day	8.8	26
Morote et al. [40]	2020	Spain	Primary and secondary	L/student/day	7.34	14

.

5. Prediction of Heating and Electrical Energy Consumption in Schools

Predicting energy consumption in schools is vital for efficient resource management, cost savings, and environmental sustainability [46]. Accurate predictions enable schools to optimize energy usage, reduce their carbon footprint, and allocate resources effectively, fostering a culture of sustainability and responsible energy management within the educational community [47]. Additionally, they aid in regulatory compliance, infrastructure planning, and ensuring operational resilience during emergencies.

Capozzoli et al. examined the amount of energy used for heating in 80 northern Italian school buildings. To estimate the energy usage, they created and assessed two models: a classification and regression tree (CART) model and a multiple linear regression (MLR) model. Both models were evaluated based on statistical coefficients. The investigation concluded that the gross heated volume, heat transfer surface area, boiler size, and window thermal conductivity primarily influenced the heating energy consumption in the school buildings analyzed [48]. An empirical study by Beusker et al. examined the variables affecting the amount of energy used for heating in urban schools and sports facilities. A random sample of 105 buildings in Stuttgart, Germany, served as the basis for the study. Various linear and nonlinear regression models were carefully formed and assessed to forecast the heating energy consumption. The authors stressed that the suggested model exhibited an excellent accuracy and satisfied every requirement for a successful evaluation [30]. Mohammed et al. proposed a model based on multiple regression to estimate the energy consumption in school buildings in Saudi Arabia. The authors highlighted that the model offers a practical and cost-effective approach that government institutions can use to assess energy consumption [22]. Alshibani conducted a study in the Eastern Province of Saudi Arabia to determine factors influencing energy consumption in school buildings. A total of 352 real energy consumption datasets from schools were used. The developed energy consumption estimation model included eleven parameters affecting the energy usage in built schools, which were used as the input variables for model development. The model was validated on eight real cases and showed an accuracy of 87.5%. The study identified the “air conditioning capacity” as the most influential factor, followed by the “total roof area of the school” [49]. Ding et al. presented a methodology that enables the prediction of annual energy consumption patterns on an hourly basis. The authors noted that specific load profiles can accurately reflect the modern energy demands of Nordic schools, and the techniques used can be extended to other types of buildings [47].

Cao et al. proposed a model for estimating the energy consumption in educational institutions by combining geographical variables and time-series data. The validity of the proposed model was confirmed by applying it to an educational institution located in Xi’an, Shaanxi Province. The results show that the integrated energy consumption estimation model achieved a reduction in the root mean square error (RMSE) values ranging from 13.64% to 34.55% compared to previous predictive models. Additionally, the mean absolute error (MAE) decreased by 10.25% to 30.54%, indicating an improved forecasting accuracy [50]. Shahid et al. developed a predictive model to estimate the electricity and thermal energy consumption. They employed sophisticated machine learning methods like autoencoders (AE), LSTM, convolutional neural networks (CNNs), and recurrent neural networks (RNNs). Real consumption data from six public schools in a Swedish municipality were used to train the model. According to the experimental results, the model’s accuracy was high, with the RMSE and the normalized RMSE (nRMSE) values between 18% and 25% for electricity, and from 20% to 30% RMSE and 5% nRMSE for thermal energy [51]. Doiphode and Najafi suggested estimating the monthly energy usage of K–12 schools in Brevard County, Florida, USA, using a multi-layer perceptron (MLP) neural network. The output variable was the monthly energy usage, while the input factors were the population, the number of working days per month, the building area, the average monthly outdoor temperature, and the relative humidity. Three years’ worth of energy usage data from 25 middle and elementary schools were used to successfully train the chosen network [52]. A study by Jurišević et al. explored the use of predictive models to identify key factors influencing heat consumption and to forecast energy use in 11 public kindergartens in Kragujevac, Serbia. Two linear models (simple and multiple linear regression) and two non-linear models (decision tree and artificial neural network) were applied. The artificial neural network proved to be the most accurate, but also the most complex to develop, while simple linear regression was the easiest to implement, but the least precise [53].

Soares Geraldi et al. presented a predictive model developed using Bayesian networks to estimate energy consumption. Over the course of three years, the authors gathered monthly power invoices from 90 public schools in southern Brazil. They also gathered information on the area of each school, the number of students, the education level, the number of floors, and the frequency of events held in the building. As potential directions for future research, the authors suggest improving the database by including more characteristics [54]. Run et al. applied an MLR model to predict the hourly electricity consumption in school buildings in southern France during the winter season. The analysis showed that the coefficient of determination (R²) for the training dataset was 74%, while for the validation dataset, it was 77% [55]. The impact of variables like the school size and the air conditioning capacity on the annual consumption was highlighted in a study by Tariq et al. that examined several artificial intelligence models, including decision trees, k-nearest neighbors, GBR, and LSTM, for forecasting the electricity consumption in schools. Their findings showed that k-nearest neighbors suffered from overfitting, whereas decision trees performed well during the training phase with low prediction errors. Diverse data ranges were handled remarkably well by the GBR and LSTM models [56]. A summary of previous research regarding heating and electrical energy consumption predictions in schools is provided in

Table 3. Summary of the previous research regarding heating and electrical energy consumption predictions in schools.

Reference	Year	Country	Type of School	Method Used for Model Development	Type of Energy	Sample Size	Performance Metrics	Key Findings
Beusker et al. [30]	2012	Germany	Elementary and high with sports field	Linear and non-linear regression	Heating	105	R2 = 0.60, MAPE = 17% (train), MAPE = 10% (validation)	Developed models to forecast heating energy consumption; emphasized model’s high accuracy and suitability for evaluation.
Capozzoli et al. [48]	2015	Italy	Not specified	MLR and CART	Heating	80—training	MAE = 108/102 MWh, RMSE = 145/142 MWh, MAPE = 15%/14%	Identified gross heated volume, heat transfer surface area, boiler size, and window thermal conductivity as primary influencers of heating energy consumption.
						5—validation
Soares Geraldi et al. [54]	2019	Brazil	Elementary and high	Bayesian network	Electrical	90	NRMSE range: 2.82–83.5% depending on model and metric basis	Suggested enhancing the database with additional characteristics for improved modeling.
Alshibani [49]	2020	Saudi Arabia	Elementary, middle, and high	ANN	Electrical	352	Accuracy = 87.5%	Validated on 8 cases; identified air conditioning capacity and total roof area as significant factors.
Doiphode and Najafi [52]	2020	Florida, USA	Elementary and high	MLP	Heating and electrical	25	Not explicitly reported	Successfully trained a neural network using three years of data to estimate monthly energy usage.
Mohammed et al. [22]	2021	Saudi Arabia	Not specified	Regression	Heating and electrical	350—training	Accuracy > 90%	Proposed a practical and cost-effective model for estimating energy consumption in Saudi Arabian schools.
						35—validation
Ding et al. [47]	2021	Norway	Not specified	MLR	Electrical	40	Electricity (best year): MAPE = 6.6%, NMBE = −0.3%, CV(RMSE) = 10.6–16.0%; Heating (DH): MAPE = 20.2–29.6%	Presented methodology for predicting annual energy consumption patterns on an hourly basis; adaptable to other building types.
Cao et al. [50]	2023	China	Complex of elementary and high	SHAP	Electrical	1	RMSE reduction: 13.64–34.55%; MAE reduction: 10.25–30.54%	Demonstrated improved forecasting accuracy over previous models.
Run et al. [55]	2023	France	Elementary and high	MLR	Electrical	9	R2: 74% (training), 77% (validation)	Highlighted the impact of variables like school size and air conditioning capacity on annual consumption.
Shahid et al. [51]	2023	Sweden	Elementary and high	RNN, LSTM, CNN, and AE	Heating and electrical	6	Electricity: RMSE/nRMSE = 18–25%; Thermal: RMSE = 20–30%, nRMSE = 5%	High accuracy in predicting energy consumption using advanced machine learning techniques.
Tariq et al. [56]	2024	Saudi Arabia	Elementary and high	Decision tree, k-nearest neighbors, GBR, and LSTM	Electrical	352	Best model (GBR): RMSE = 3559.43, MAE = 1510.67, MAPE = 4.14%, R2 = 0.9995; LSTM (testing): MAPE = 6.29%, R2 = 0.9757	Observed overfitting in KNN; decision trees performed well during training with low prediction errors; GBR and LSTM handled diverse data ranges effectively.

.

Due to the large number of analyzed studies regarding the prediction of heating and electrical energy consumption in schools, a graphical representation of the published analyzed studies regarding the period of publication and a graph on the number of studies with regards to the models used in the study are provided in Figure 1 and Figure 2.

6. Prediction of Water Consumption in Educational Buildings

Despite the growing emphasis on sustainability and resource efficiency in educational environments, the academic literature remains notably scarce in addressing the specific issue of water consumption in schools. While energy consumption in educational buildings has been widely studied, research focusing on water usage is limited and fragmented, with few studies explicitly targeting primary or secondary school facilities. This lack of data and analysis presents a challenge for the development of comprehensive water management strategies in the education sector. In light of this research gap, the following subsection presents relevant studies conducted in broader educational settings, such as university campuses and preschools. Although these are not directly school-based, they offer valuable insights and methodological approaches that may be applicable or adaptable to school environments.

Almeida et al. conducted a study aimed at estimating the energy and water consumption on the Paricarana campus of the Federal University of Roraima in Brazil. Their research methodology included measuring the time required to fill one liter of water as a means of determining the water flow rates from various plumbing fixtures. Additionally, interviews with building administrators were conducted to identify the types of activities performed within individual rooms and buildings, thereby enabling the estimation of operating times for lighting and cooling equipment. To assess the consumption habits of the academic population, indirect questionnaires were distributed via email in June 2017. The study also incorporated direct surveys to evaluate water usage for the purpose of room cleaning [57].

Jurišević et al. analyzed the water consumption in 13 preschool facilities located in the city of Kragujevac, Serbia, over a three-year period. The authors identified 21 parameters influencing water usage and developed six predictive models to estimate the consumption levels. Their findings indicated that the random forest (RF) algorithm achieved the highest overall performance. Furthermore, MLR demonstrated a comparable accuracy to RF for buildings with a monthly water consumption exceeding 200 m³. Both approaches yielded satisfactory results, offering end users the option to prioritize either the model performance (RF) or the methodological simplicity (MLR) [58]. A summary of previous research regarding water consumption predictions in educational buildings is presented in

Table 4. Summary of the previous research regarding water consumption predictions in educational buildings.

Reference	Year	Country	Type of Educational Building	Method Used for Model Development	Sample Size	Performance Metrics	Key Findings
Almeida et al. [57]	2021	Brasil	Campus	Volume and time measurement approach	1	Not reported (qualitative only)	Combined water/energy estimation via surveys and usage observations. Focused on operational habits.
Jurišević et al. [58]	2021	Serbia	Preschool	MLR and RF	13	MAPE: RF = 14%, DT = 15%; R2: RF = 0.89, DT = 0.90	A total of 21 variables analyzed. RF offered best performance and robustness; MLR simpler, but sufficient when water use exceeded 200 m3/month.

.

Although the literature is limited, the reviewed studies suggest a preliminary framework for water consumption predictions in educational buildings. Common predictive variables include the student population, building size, number of restrooms, cleaning frequency, fixture types, and operational schedules. The two identified studies used different methodological approaches. Almeida et al. [57] employed direct flow measurements and surveys, while Jurišević et al. [58] applied machine learning and regression techniques. Among the models tested, the random forest algorithm demonstrated the highest performance, particularly in buildings with large-scale consumption, while multiple linear regression offered a simpler and more interpretable alternative with reasonably accurate results. Based on these findings, the authors proposed a feasible modeling strategy for school water consumption predictions that combines easily accessible data (e.g., the number of students, area, and volume) with machine learning methods such as ANN or RF, which would allow for a high accuracy while maintaining flexibility in the model complexity based on data availability. Additionally, the gradual adoption of smart water metering technologies could enable real-time data collection and support the development of more robust, generalizable models tailored to primary and secondary schools.

7. Discussion

The analysis presented in this study underscores the significant role that energy and water consumption play in the operational and environmental performance of school infrastructure. By reviewing previous research, it becomes evident that educational buildings are substantial consumers of energy and water resources, with their consumption patterns influenced by a variety of factors, including the geographical location, climate, building age, occupancy levels, and school type.

The findings regarding energy consumption reveal several key insights. Notably, there is considerable disparity in energy usage across schools globally. These differences are driven by climate conditions, national energy policies, the adoption of energy-efficient technologies, and the architectural design of school buildings. Heating, lighting, and air conditioning systems are typically the primary contributors to the overall energy demand in schools. As such, strategies aimed at reducing consumption should prioritize improvements in building insulation, the modernization of HVAC systems, and the integration of energy-efficient lighting and controls.

In contrast, water consumption in schools has received far less attention in academic research. While water usage is a fundamental component of school operation, related to sanitation, hygiene, cleaning, and, in some cases, landscaping, the literature on its analysis and prediction is notably limited. The present study found that only two papers have addressed the prediction of water consumption in educational buildings, and neither was specific to schools. These studies instead examined broader educational settings, such as university campuses or preschools. As a result, to the best of the author’s knowledge, there is currently no published research focusing specifically on water consumption prediction models in primary or secondary schools, representing a significant gap in the field.

The available findings suggest that water consumption is influenced by several building-level and behavioral factors, including the student population, the fixture efficiency, and maintenance practices. Efforts to improve the water efficiency in schools should therefore prioritize the installation of low-flow fixtures, leak detection, maintenance programs, and behavioral interventions aimed at reducing unnecessary usage.

While this review confirms the scarcity of research on water consumption in school buildings, it is important to consider the underlying reasons for this limitation. Data collection efforts are often burdened by inconsistent or outdated metering infrastructure, particularly in older schools or in regions with limited technical capacity. In addition, variations in school types (e.g., preschool, primary, secondary), occupancy schedules, and facility use make it difficult to develop standardized metrics. Administrative obstacles, such as limited access to utility records and low response rates to questionnaires, as noted by Morote et al. [40], also further constrain data availability.

In addition to a descriptive analysis, the application of predictive modeling for energy and water use presents valuable opportunities for sustainable building management. Regression techniques, ensemble learning models, and artificial neural networks have demonstrated strong potential in forecasting the resource demand and identifying inefficiencies. However, the integration of these models in school settings remains uneven. The scarcity of data, particularly on water usage, continues to limit the broader adoption and development of accurate, school-specific models.

This study contributes to the theoretical understanding of sustainability in educational infrastructure by integrating a dual focus on both energy and water consumption, with particular attention to the role of predictive modeling. While prior studies such as those by Wang [25] and Kim et al. [24] have highlighted disparities in energy use across school levels and regions, this review extends the discussion by drawing attention to the underexplored area of water consumption and its modeling. Furthermore, this paper synthesizes evidence from studies using various predictive approaches—including MLR, decision trees, and neural networks—such as those by Jurišević et al. [53], Beusker et al. [30], and Shahid et al. [51], revealing patterns in model complexity, accuracy, and usability. This paper thus offers a comparative perspective on the modeling performance and suitability in school contexts.

Additionally, government policies and financial incentives play a vital role in shaping energy and water consumption practices in educational buildings. Regulations related to energy performance standards, mandatory audits, or building codes can significantly influence how schools manage their resource use and prioritize upgrades. Additionally, access to public funding or incentive programs, such as grants for energy-efficient renovations or smart metering installations, can enable schools, particularly in low-income regions, to adopt predictive modeling tools and implement conservation measures. In many of the reviewed studies, policy frameworks were either absent or underreported, which may have contributed to regional disparities in data availability and efficiency outcomes. Integrating predictive models into policy-supported programs could enhance the adoption and long-term impact, especially when paired with targeted financial mechanisms.

Furthermore, beyond technical solutions and policy support, the successful adoption of resource efficiency measures in schools also depends on the active involvement of key stakeholders, including school administrators, teachers, and students. Administrators are essential for overseeing implementation, allocating resources, and maintaining systems, while teachers can integrate sustainability practices into the curricula and promote behavioral changes. Students, as daily users of school facilities, play a crucial role in shaping consumption habits and can provide change through awareness campaigns and engagement initiatives. Building a culture of shared responsibility not only improves the effectiveness of technical interventions, but also ensures their long-term sustainability. Therefore, future efforts to enhance the energy and water efficiency in schools should include educational and participatory components for all stakeholders.

8. Conclusions

The findings of this study emphasize the need for systematic approaches to analyzing and predicting the energy and water consumption in school environments. Given their high occupancy and continuous operation, schools are ideal candidates for implementing sustainability measures that can lead to significant environmental and economic benefits.

Energy consumption has been more widely studied, with multiple predictive models showing promise for improving efficiency and planning. However, water consumption remains an underrepresented area, especially in predictive research. The fact that only two identified studies have addressed water consumption forecasting, and neither in school buildings, underscores an important research gap. Addressing this void is critical for advancing sustainability efforts across the education sector.

Future research should prioritize the development of robust, school-specific models for water consumption forecasting, with a focus on primary and secondary school buildings across different geographic and socio-economic contexts. The methodological paths should include both statistical and machine learning techniques, such as multiple linear regression and ensemble models, applied to real-world datasets. To support model development, future studies should use diverse data acquisition methods, including smart metering technologies, structured surveys, utility billing records, and building-level audits. Collaboration with local education authorities should be encouraged in order to access relevant data and conduct pilot studies.

While predictive modelling of the energy consumption in school buildings has advanced significantly, research on water consumption remains limited and underdeveloped. As reflected in the structure of this review, the authors have addressed these areas separately due to the current lack of predictive modelling studies focused specifically on water use in primary and secondary schools. However, the growing emphasis on holistic sustainability underscores the importance of integrated approaches. Future research should explore the development of combined models for energy and water consumption. Such integration could support more comprehensive resource management strategies in educational institutions and address emerging challenges related to both environmental performance and operational resilience.

In summary, the integration of consumption analyses and predictive modelling provides a strategic path forward for enhancing the sustainability of educational buildings. Expanding this body of knowledge, especially in under-researched areas such as school water usage, will be essential for shaping future policies, designs, and operational practices in the education sector.