An Explainable Model for Waste Cost Prediction: A Study on Linked

Open Data in Italy

Lerina Aversano

1 a

, Martina Iammarino

2 b

, Antonella Madau

3 c

, Debora Montano

1,4 d

and Chiara Verdone

Department of Agricultural Sciences, Food, Natural Resources and Engineering, University of Foggia, Foggia, Italy

Department of Information Science and Technology, Pegaso University, Naples, Italy

Department of Engineering, University of Sannio, Benevento, Italy

Department of Surgery, Medicine, Dentistry and Morphological Sciences with Interest in Transplantology, Oncology and

Regenerative Medicine, University of Modena and Reggio Emilia, Italy

Keywords:

Open-Data, Machine Learning, Explainability, XAI, SHAP.

Abstract:

Artiﬁcial intelligence and machine learning models are emerging as essential tools for optimizing municipal

solid waste management and supporting policy decisions. However, transparency and interpretability of these

models’ predictions continue to be major obstacles. Recent advances in Explainable Artiﬁcial Intelligence

(XAI) techniques have made it possible to explain speciﬁc model decisions and guarantee that the outcomes

are intelligible and useful. Using high-quality Italian open data in the form of Linked Open Data (LOD), this

study investigates the beneﬁts and viability of creating explainable models in italian municipalities. To achieve

this, a method for using connected and open statistical data to create explainable models is provided. Addi-

tionally, a case study is presented, covering four years, in which waste management expenses are predicted

and interpreted using connected data about Italian municipalities, categorizing them into three cost bands.

CatBoost was selected as the predictive model’s algorithm, and the SHAP framework was used to guarantee

the predictions’ transparency. Through transparent and accountable data management, this effort seeks to il-

lustrate how cutting-edge technologies can enhance the sustainability of public programs.

1 INTRODUCTION

Artiﬁcial intelligence (AI) is revolutionizing many

industries, demonstrating signiﬁcant potential in op-

timizing processes, reducing costs, and increasing

decision-making effectiveness (Rashid and Kausik,

2024). The main application areas are healthcare

(Ardimento et al., 2023), transportation (Jevinger

et al., 2024), industry (Aversano et al., 2023), edu-

cation, and public administration (Kalampokis et al.,

2021), each of which beneﬁts from AI’s ability to ana-

lyze large amounts of data, identify complex patterns,

and provide accurate predictions. In public adminis-

tration, AI drives innovation to improve public ser-

vices, optimize resource management, and increase

https://orcid.org/0000-0003-2436-6835

https://orcid.org/0000-0001-8025-733X

https://orcid.org/0009-0003-2227-9778

https://orcid.org/0000-0002-5598-0822

transparency. Predictive models are used to prevent

fraud, improve public safety, and optimize adminis-

trative processes.

On the other hand, open data is a key resource

for innovation and digital transformation, thanks to

its ability to make large amounts of information pro-

duced by public institutions, research institutions, and

private organizations accessible and reusable (Park

and Gil-Garcia, 2022). These data, made available

in open and standardized formats, cover a wide range

of areas, including demography, economics, environ-

ment, healthcare, and transportation. Their diffusion

promotes transparency, citizen participation, and the

development of innovative solutions to address social,

economic, and environmental challenges.

The integration of open data and AI offers ex-

traordinary opportunities to extract value from this in-

formation, transforming it into useful knowledge to

support informed decisions and policies (Wani et al.,

2024). Open data provides an essential basis for

446

Aversano, L., Iammarino, M., Madau, A., Montano, D. and Verdone, C.

An Explainable Model for Waste Cost Prediction: A Study on Linked Open Data in Italy.

DOI: 10.5220/0013650800003964

In Proceedings of the 20th International Conference on Software Technologies (ICSOFT 2025), pages 446-453

ISBN: 978-989-758-757-3; ISSN: 2184-2833

training machine learning models, which require large

amounts of data to identify patterns and generate ac-

curate predictions. Through the use of advanced tech-

nologies, such as Linked Open Data, it is possible to

combine datasets from different sources, creating in-

tegrated information ecosystems that enable new AI

applications.

Explainable artiﬁcial intelligence (XAI) adds a

further level of value to the integration of open data

and artiﬁcial intelligence, making machine learning

models transparent and interpretable. Through the

use of techniques such as SHAP (Shapley Additive

exPlanations) or LIME (Local Interpretable Model-

agnostic Explanations), it is possible to understand

the contribution of individual factors to model predic-

tions (Dwivedi et al., 2023). This is particularly rele-

vant in contexts such as public administration, where

transparency and trust in decision-making processes

are essential to promote the adoption of innovative

AI-based solutions.

In this study, we explore the creation of XAI mod-

els based on open and linked statistical data, apply-

ing them to the prediction of waste management costs

in Italian municipalities over a period of four years.

The analyzed data include socio-demographic, geo-

graphic, and economic information. Costs were clas-

siﬁed into three bands, and the best model was se-

lected based on the area under the ROC (Receiver Op-

erating Characteristic) curve. Finally, to interpret the

results, the SHAP framework was used to identify the

determining factors for each prediction.

The main objective of this work is to assess the

value and feasibility of using high-quality open gov-

ernment data for the creation of explainable models.

Through the case study on waste costs, we propose

a structured process to exploit open-linked statistical

data, highlighting the potential of these technologies

in supporting informed decisions and more effective

policies.

The rest of the document is divided as follows:

Section 2 reports related works; Section 3 details the

approach used for the case study; Section 4 describes

the characteristics of the experiments conducted, Sec-

tion 5 discusses the results obtained, and Section 6

reports the conclusions.

2 RELATED WORKS

Due to its potential to improve decision-making pro-

cesses and optimize operations across various indus-

tries, integrating AI with open data has garnered in-

creasing attention in recent years. The collected and

standardized data, in particular, provide a crucial re-

source for the demonstration of machine learning

(ML) models and enable the demonstration of useful

insights from complex datasets.

In particular, the use of Linked Open Data (LOD)

has revolutionized the ability to combine information

from heterogeneous sources to create integrated in-

formation ecosystems. Several studies have explored

how LOD can be used to support predictive models

in complex contexts, such as public resource manage-

ment and urban planning (Lnenicka and Nikiforova,

2021). The availability of high-quality open data is

essential to improve the effectiveness of predictive

models and promote data-driven solutions. The au-

thors in (Corea et al., 2023) discuss principle-based

approaches to AI adoption and their impact on the

use of open data in Italy and the EU; while, in (Orusa

et al., 2024) the use of satellite data and geospatial

deep learning techniques to support the distribution

of ﬁnancial contributions to European municipalities

is explored, proposing an innovative system for public

administrations, which exploits geomatics and remote

sensing to improve resource allocation based on spa-

tial data and advanced technologies.

Therefore, although numerous studies have ex-

plored the potential of artiﬁcial intelligence applied to

LODs to improve processes and policies in public ad-

ministration, a growing consensus emerges on the im-

portance of ensuring transparency and interpretability

in the models used. In this regard, numerous studies

focus on the use of Explainable AI (XAI) for LODs in

public administration. Among these, for example, the

study (Kalampokis et al., 2021) examines the appli-

cation of XAI techniques on linked open data, high-

lighting their potential to improve public services and

policies.

Papadakis et al. (Papadakis et al., 2024) explore

how XAI can support transparency and trust in public

sector decision-making, presenting a reference archi-

tecture for AI-based policy development, virtualized

tools for data-driven policy speciﬁcation and imple-

mentation, and a machine learning framework that en-

ables the creation of transparent and explainable mod-

els. (Karamanou et al., 2022) focuses on the creation

of a predictive model that uses machine learning to

identify variables that inﬂuence house prices. The

data used comes from government sources, in par-

ticular the Scottish Statistics Portal. XAI techniques

were applied to make the model transparent and in-

terpretable, allowing users, including policymakers

and citizens, to understand the drivers of the fore-

casts. Our study applies XAI not only to motivate

forecasts, but also to directly drive operational deci-

sions on waste management costs.

Similar to this study, our goal is to develop an

An Explainable Model for Waste Cost Prediction: A Study on Linked Open Data in Italy

447

equally reliable and transparent model but applied to

the context of waste management. In particular, we

intend to use high-quality linked open data, relating

to socio-demographic, administrative, and geograph-

ical information, to build a predictive system capable

of estimating waste management costs in Italian mu-

nicipalities.

In literature, waste cost forecasting is a topic that

has already been addressed. In (Fasano et al., 2021)

a deep learning approach is presented that allows to

precisely identify the critical factors that inﬂuence

waste management, providing useful information to

plan more aware interventions and to promote the

transition to a circular economy, including the number

of rooms in residential homes and the year of con-

struction of buildings on waste production. Finally,

Roseck

y et al. in (Roseck

y et al., 2021) presented

models capable of predicting with good accuracy the

amount of waste generated in speciﬁc areas and esti-

mating the percentage of separate and non-separated

collection, identifying demographic variables, eco-

nomic factors, and territorial characteristics as key

factors. These two studies focus mainly on quantita-

tive predictor variables, while our approach also inte-

grates socio-demographic and administrative factors,

offering a qualitative explanation of costs.

Our contribution to the ﬁeld is the use of XAI ap-

proaches into municipal solid waste management pre-

diction models. While prior research has mostly con-

centrated on creating models to predict waste genera-

tion and optimize management strategies, our method

seeks to address one of the primary drawbacks of

these studies: the results’ lack of transparency and

interpretability. Our work attempts to give thorough

explanations of the elements inﬂuencing model pre-

dictions using XAI tools like SHAP. In addition to

increasing public administrators’ and stakeholders’

conﬁdence in the outcomes, this makes it possible to

pinpoint the crucial factors that should be addressed

in order to maximize management tactics.

3 APPROACH

This section describes the approach followed to carry

out the experiments, covering the data used, the clas-

siﬁers, and the explainability method applied.

3.1 Data

To provide a comprehensive overview of Italian mu-

nicipalities, we utilized real data on waste costs (mea-

sured in kilograms per inhabitant) and waste man-

agement practices. Our analysis focused on four key

Table 1: Distribution of features.

Group Number of features

Waste Cost 11

Waste management 20

Socio-demographic 8

Income 38

Geographic Information 8

areas: Waste cost and cost distribution, Waste man-

agement and collection, including separate waste col-

lection, Socio-demographic data such as population

size and gender distribution, Income data reﬂecting

the economic status of citizens and Geographic infor-

mation regarding the municipalities locations.

The data were integrated to offer a detailed and

comprehensive representation of each municipality’s

waste management landscape. Data span the years

2019, 2020, 2021, and 2022, including a total of 6,437

municipalities, representing a signiﬁcant percentage

of the total of 7,904 Italian municipalities

. All data

is publicly accessible.

The dataset comprised 19394 instances, with 85

features categorized into 5 groups, as illustrated in Ta-

ble 1.

The information relating to waste costs comes

from the Waste Registry, organized in a National Sec-

tion at the Higher Institute for Environmental Protec-

tion and Research (ISPRA

) and Regional Sections

or the Autonomous Provinces of Trento and Bolzano,

coordinated by the Regional Agencies for Environ-

mental Protection and the Autonomous Provinces. IS-

PRA has implemented the National Section through

an advanced IT system, known as the Telematic

Catasto, which provides a complete, constantly up-

dated, and easily consultable knowledge framework

regarding waste management.

The analyzed data outlines the management of ur-

ban waste and the total cost per capita, calculated

based on the following components: CRT(Costs of

collection and transportation of unsorted urban waste,

CTS (Costs of treatment and disposal of unsorted

urban waste), CRD (Costs of collection and trans-

portation of sorted urban waste),CTR(Costs of treat-

ment and recycling of sorted urban waste), CSL(Costs

of street sweeping and washing) and CC (Common

costs), which include: CARC - Operating costs for the

management of tariffs and relations with users, CGG

- Management costs relating to both personnel not

directly employed and the share of structural costs,

CCD - Costs relating to the share of bad debts and

COAL - Includes the share of operating costs of the

territorially competent bodies, of ARERA and local

//www.istat.it/storage/ASI/2022/Sintesi.pdf

//www.catasto-riﬁuti.isprambiente.it/index.php?pg=

ICSOFT 2025 - 20th International Conference on Software Technologies

448

charges. For the waste management group, the con-

sumption values in tons are based on the cadastral ori-

gin and consider differentiated collection types, such

as Glass, Aluminum, Plastic, etc.

The assessment of income characteristics included

both the source of income—such as pensions, self-

employment, or rental income from properties—and

the level of income, which was categorized into var-

ious bands ranging from less than 10,000 euros to

over 120,000 euros. This approach provides a com-

prehensive overview of the economic situation of cit-

izens. The data was taken from the website of the

Italian government, in particular from the Ministry of

Finance

The analysis included not only cost data but also

detailed socio-demographic information. Speciﬁ-

cally, it examined the demographic composition, fo-

cusing on the distribution of the population by gen-

der, with particular attention to the number of men

and women living in each municipality. Additionally,

the population was analyzed by age groups, catego-

rized into four main intervals: young people (up to 25

years), adults (26 to 50 years), middle-aged individu-

als (51 to 75 years), and the elderly (over 75 years).

In parallel, geographical information

was col-

lected regarding the surface area and population den-

sity of each municipality. Furthermore, the geograph-

ical position was assessed in terms of the munici-

pality’s belonging to the Italian macro-areas: North,

Center, or South, and whether or not the municipality

belongs to the group of small municipalities

, con-

tributing to a complete and detailed representation of

the territorial and demographic characteristics.

The variable used for classiﬁcation is the total

cost of waste per inhabitant, which has been cate-

gorized into three distinct levels: low, medium, and

high costs. Based on all the information previously

described, the ﬁnal dataset comprises a total of 85 fea-

tures and 19394 instances.

3.2 Predictive Model

The study evaluated several machine learning algo-

rithms to identify the best-performing model based

on the area under the ROC curve. The classiﬁers

tested included Random Forest, Extra Trees, XG-

Boost, Decision Tree, CatBoost, AdaBoost, and Gra-

dient Boosting. Each uses different ensemble or

boosting techniques to improve predictive accuracy,

with some (like CatBoost and XGBoost) offering ad-

https://www1.ﬁnanze.gov.it/ﬁnanze/analisi stat/

public/index.php?opendata=yes

https://esploradati.istat.it/databrowser/

//www.anci.it/atlante-dei-piccoli-comuni/

vanced features such as handling categorical data and

regularization.

These different algorithms provide a comprehen-

sive set to tackle complex classiﬁcation tasks by lever-

aging unique strengths and addressing different chal-

lenges in data modeling and forecasting.

3.3 Explainability

Explainable artiﬁcial intelligence (XAI) is an emerg-

ing ﬁeld in machine learning that aims to make AI

models more interpretable and understandable for hu-

mans. The primary objective is to open the ”black

box” nature of AI by providing transparent expla-

nations of how these models operate. This trans-

parency not only facilitates a better understanding of

the decision-making process but also aids in feature

selection by identifying which attributes signiﬁcantly

inﬂuence predictions. This approach clariﬁes how

each feature, in addition to those related to costs, con-

tributes to the target variable, providing a clear picture

of their respective impacts.

In this study, we utilized the Shapley Additive Ex-

planations (SHAP) method which offers a theoreti-

cal framework for interpreting the logic behind model

predictions. SHAP calculates the average marginal

contribution of each feature across all possible fea-

ture combinations, providing a detailed overview of

the role each attribute plays for every instance in the

dataset. By approximating the impact of a feature on

a model’s output, SHAP enables explanations of pre-

dictions without the need to rerun the model on all

combinations of features.

The method involves calculating Shapley values

for each input feature, which are then applied to every

instance in the dataset. This allows for both a granular

analysis of individual instances and a broader under-

standing of feature behavior. Shapley values ensure

an equitable distribution of prediction contributions

across all features.

Their speciﬁc mathematical formulation is:

g(z

′

) = φ

∑

′

(1)

in which g represents the explanatory model,

′

∈

{

0, 1

}

make up the coalition vector, where M is the maxi-

mum coalition size. Additionally,

∈ R

represents the feature attribution for a given feature j,

denoting the Shapley values.

These values indicate the individual contribution

of each input variable to the model’s predictions for

An Explainable Model for Waste Cost Prediction: A Study on Linked Open Data in Italy

449

speciﬁc examples in the test set. This analysis is cru-

cial for understanding not only which variables inﬂu-

ence the forecasts but also the extent of their inﬂu-

ence. In other words, SHAP decomposes the complex

predictions of the model, allowing us to assign spe-

ciﬁc importance to each variable in generating every

model prediction.

4 EXPERIMENTAL SETTINGS

This section describes the experimental phases con-

ducted for the analysis and prediction of waste costs

in Italian municipalities. The main objective is to

build a predictive model capable of estimating the

cost of waste management per inhabitant, using socio-

demographic, economic, and geographical data, in

addition to information on collection and disposal

practices. Furthermore, the aspect of the explainabil-

ity of the model is explored in order to interpret the

results obtained and identify the factors that mostly

inﬂuence the cost of waste management.

4.1 Prediction of Waste Cost

For the prediction of waste costs, the Hold-out vali-

dation method was used, where the data was split in a

70/30 ratio for training and testing the model, respec-

tively. In this approach, 70% of the data is used to

train the model, while the remaining 30% is reserved

for testing, allowing the model’s performance to be

evaluated on unseen data during the training process.

The metrics used for analyzing the model’s per-

formance include accuracy, precision, recall, F-score,

and ROC (Receiver Operating Characteristic curve).

Accuracy measures the percentage of correct predic-

tions compared to the total predictions made. It is use-

ful as a general indicator of performance, but might

not be sufﬁcient when the classes are imbalanced.

Precision is the ratio of true positives (correctly pre-

dicted as positive) to all positive predictions, indicat-

ing how precise the model is in predicting the posi-

tive class. Recall (or sensitivity) measures the model’s

ability to correctly identify all positive instances, in-

dicating whether the model is capturing all possible

positives. F-score is the harmonic mean of precision

and recall, combining both indicators into a single

measure that balances the two aspects, useful when

classes are imbalanced. ROC curve represents the re-

lationship between the true positive rate (TPR) and

the false positive rate (FPR), providing a useful vi-

sualization for comparing the performance of differ-

ent models, especially in imbalanced class scenarios.

The area under the curve (AUC-ROC) is an important

metric for evaluating the overall ability of the model

to distinguish between the classes.

Finally, the model with the best performance was

chosen based on these metrics, with particular atten-

tion to the ROC, as it provides a more comprehensive

view of the model’s behavior in the presence of imbal-

anced classes. The ROC curve allows us to see how

the trade-off between true positives and false positives

changes as the classiﬁcation threshold varies, making

it a crucial tool for optimizing model selection. Sub-

sequently, this model was made explainable to bet-

ter understand how the variables inﬂuence predictions

and ensure transparency in the model’s decisions.

4.2 Explainability of the Model

A model interpretability study was conducted to ana-

lyze how various characteristics inﬂuence the predic-

tions related to waste costs. To achieve this, the SHAP

library was utilized, which assigns a ”contribution”

value to each feature for every prediction made by the

model. The SHAP algorithm operates by analyzing

the model’s predictions based on the input data. It

calculates the contribution of each variable, tracking

how each feature impacts the ﬁnal result and provid-

ing a clear, interpretable view of the decision-making

process. Initially, the importance of all features in the

dataset was evaluated using a SHAP bar plot. This vi-

sual representation provides a straightforward way to

assess the relative importance of each variable, sorted

by its overall impact on the model.

Next, variables of the costs were removed to

examine how additional information, such as so-

ciodemographic, geographic, income, and waste

collection-related data, inﬂuenced the results.

SHAP plots were used to visually analyze the im-

pact of variables on the model. The SHAP Sum-

mary Plot highlighted the most inﬂuential features

by showing the distribution and importance of SHAP

values, while the SHAP Dependence Plot illustrated

how changes in individual feature values affect pre-

dictions, offering insights into variable relationships

and interactions inﬂuencing waste management costs.

These tools provided a comprehensive under-

standing of how various socio-demographic, geo-

graphic, income, and waste collection-related vari-

ables inﬂuence waste costs. This knowledge enables

the identiﬁcation of key factors and the optimization

of the model to enhance both its predictive accuracy

and interpretability.

ICSOFT 2025 - 20th International Conference on Software Technologies

450

5 RESULTS AND DISCUSSION

This section presents and analyzes the results of

applying the predictive model to waste manage-

ment costs in Italian municipalities. It evaluates the

model’s accuracy, the inﬂuence of various features,

and the potential policy implications. The explain-

ability aspect is also explored to understand the con-

tribution of each factor to the cost and to identify crit-

ical issues and areas for improvement. Speciﬁcally,

Table 2 shows the baseline results using all initial fea-

tures, with seven classiﬁers evaluated based on Accu-

racy, Precision, Recall, F-Score, and ROC-AUC met-

rics.

The results indicate a solid overall performance,

with CatBoostClassiﬁer emerging as the best classi-

ﬁer. The following metrics were achieved: Accuracy

= 95%, Precision = 95%, Recall = 85%, F-score =

90%, and ROC = 0.99. These results demonstrate the

model’s high predictive power and optimal separation

between classes. XGBClassiﬁer comes close with an

accuracy of 92% and a ROC-AUC score of 0.98. In

contrast, AdaBoostClassiﬁer exhibits the lowest per-

formance, with an accuracy of 72% and the same F1-

score. Therefore, the CatBoostClassiﬁer was selected

as the primary model for explainability analysis. It

not only showed excellent performance in terms of ac-

curacy but also provided a solid basis to interpret and

analyze its predictions via techniques such as SHAP.

The outcomes of the second phase of the study,

which examines the model’s explainability, are pre-

sented in Figures1, 2.

Figure 1 illustrates a horizontal bar chart show-

ing the relative importance of the variables within the

model. Each bar represents a speciﬁc variable, with

the length of the bar reﬂecting the average impact

of the variable on the model’s predictions. Variables

with longer bars indicate a greater inﬂuence on the

outcome.

Upon analyzing the chart, it is observed that the

variables CRDab, CCab, CTSab, CRTab, and CSLab

(i.e., those related to costs) have signiﬁcantly longer

bars compared to the others. This suggests that

these variables play a crucial role in determining the

model’s predictions, emerging as the main drivers of

those predictions.

Figure 1: Bar Plot All.

Thus, the model is heavily inﬂuenced by a small

subset of variables, while the other variables, al-

though present in the dataset, have a marginal or neg-

ligible impact on the predictions.

Graph 2 illustrates the impact of the less inﬂu-

ential variables on the prediction after excluding the

most signiﬁcant ones. In this way, factors that, al-

though not the primary drivers, still play a role in in-

ﬂuencing the model’s predictions are identiﬁed. The

features ”Reddito da fabbricati - Euro,” ”Reddito da

lavoro dipendente e assimilati - Euro,” and ”Fascia 0-

25” (referring to the number of citizens aged 0 to 25)

have the highest impact among the remaining vari-

ables and show a signiﬁcant inﬂuence. This suggests

that integrated information related to income and pop-

ulation demographics contributes meaningfully to the

model’s predictions.

Last ﬁgures 3, 4,5, present SHAP summary plots

that illustrate the relative impact of various features

across the entire dataset. Each graph refers to a differ-

ent cost class: the ﬁrst represents the low-cost class,

the second the medium-cost class, and the third the

high-cost class. The variables are ordered according

to their importance, with the most inﬂuential ones po-

sitioned at the top. The color of the points indicates

the value of the variable, with shades ranging from

blue for the lowest values to red for the highest values,

Table 2: Results Baseline.

Classiﬁer Accuracy Precision Recall F-Score ROC-AUC Score

RandomForrest 0.8552 0.8557 0.8552 0.8555 0.9615221249

ExtraTrees 0.8177 0.8191 0.7673 0.818 0.9359426579

XGBClassiﬁer 0.9236 0.9245 0.9236 0.9239 0.9888886590

DecisionTree 0.7784 0.7785 0.7784 0.7783 0.8338225201

CatBoostClassiﬁer 0.9495 0.9499 0.9495 0.9496 0.9943761924

AdaBoostClassiﬁer 0.7192 0.7685 0.7192 0.7167 0.8470074004

GradientBoostingClassiﬁer 0.8868 0.8876 0.8868 0.887 0.9759861166

An Explainable Model for Waste Cost Prediction: A Study on Linked Open Data in Italy

451

Figure 2: Bar Plot without cost.

Figure 3: Summary Plot low cost

Figure 4: Summary Plot medium cost.

while the horizontal position of each point reﬂects the

associated SHAP value, determining the variable’s ef-

fect on the model’s prediction.

The analysis of the graphs shows that the income

from buildings and the income from employment in

euros are among the most determining factors for the

Figure 5: Summary Plot High cost.

prediction of the cost of waste, regardless of the class

considered. The population and the distribution of in-

come brackets signiﬁcantly inﬂuence the predictions,

suggesting that municipalities with a greater concen-

tration of citizens with medium or high incomes tend

to fall into the higher-cost classes. It is also noted that

some variables, such as income from participation and

the income of the entrepreneur in simpliﬁed account-

ing, have a signiﬁcant but variable impact depending

on the cost class considered, highlighting a possible

correlation between the type of prevalent income and

waste management expenses.

In the case of the low-cost class, the most im-

pactful variables are closely linked to the population

and the lower income bracket, suggesting that mu-

nicipalities with a prevalence of citizens with lower

incomes tend to have lower management costs. For

the medium-cost class, a greater incidence of over-

all income and distribution among age groups is ob-

served, indicating that socio-demographic factors and

economic distribution play a more marked role. Fi-

nally, for the high-cost class, the model gives greater

importance to higher incomes and the frequency of

speciﬁc sources of income, highlighting that munic-

ipalities with a greater presence of wealthy citizens

tend to have higher waste management costs.

Nonetheless, the proposed approach also presents

some limitations. While SHAP improves the inter-

pretability of the predictions, the observed correla-

tions between variables and waste management costs

should be interpreted with caution and not as direct

causal relationships.

6 CONCLUSION

The cost of waste management is a key concern for

local governments, inﬂuenced not only by collection

ICSOFT 2025 - 20th International Conference on Software Technologies

452

and disposal processes but also by socio-economic,

demographic, and geographic factors. This study in-

troduces an innovative approach that integrates open

data from ofﬁcial sources (ISTAT, Waste Registry,

Ministry of Finance) to analyze ﬁve main aspects:

waste cost per capita, percentage of separate collec-

tion, average income, socio-demographic data, and

geographical characteristics.

The study employed a two-phase methodology:

ﬁrst, it forecasted waste management costs (catego-

rized as high, medium, or low); second, it analyzed

which variables most inﬂuenced these forecasts us-

ing various machine learning models and SHAP for

explainability. Data were sourced from ofﬁcial insti-

tutions to ensure quality. The results offer practical

value for municipalities, enabling more efﬁcient re-

source allocation and tailored interventions. The ex-

plainable model also promotes transparency and data-

driven policymaking. Future efforts will aim to de-

velop a user-friendly decision support tool for public

administrators.

The model achieved excellent classiﬁcation re-

sults (AUC-ROC of 99%), conﬁrming the value of

integrating socio-economic, environmental, and ter-

ritorial data in analyzing waste management costs.

SHAP analysis identiﬁed key inﬂuencing factors such

as separate collection rates, average income, popu-

lation density, and geographic location. Notably, a

higher rate of separate collection does not always lead

to lower costs, highlighting that economic, social, and

territorial characteristics play a critical role in deter-

mining waste management expenses.

Overall, these results highlight the advantages

of integrating and analyzing open data to support

waste management policies. An approach based on

multidimensional data can allow administrations to

adopt more targeted and efﬁcient strategies, optimiz-

ing available resources and reducing costs without

compromising the quality of the service. This study

provides useful insights for future research and for the

development of decision-support tools that can help

public bodies and policymakers to improve the sus-

tainability and effectiveness of municipal waste man-

agement.

REFERENCES

Ardimento, P., Aversano, L., Bernardi, M. L., Cimitile,

M., Iammarino, M., and Verdone, C. (2023). Evo-

gunet3++: Using evolutionary algorithms to train

unet-based architectures for efﬁcient 3d lung cancer

detection. Applied Soft Computing, 144.

Aversano, L., Bernardi, M. L., Cimitile, M., Iammarino,

M., and Verdone, C. (2023). A data-aware explain-

able deep learning approach for next activity predic-

tion. Engineering Applications of Artiﬁcial Intelli-

gence, 126:106758.

Corea, F., Fossa, F., Loreggia, A., Quintarelli, S., and

Sapienza, S. (2023). A principle-based approach to

ai: the case for european union and italy. Ai & Soci-

ety, 38(2):521–535.

Dwivedi, R., Dave, D., Naik, H., Singhal, S., Omer, R.,

Patel, P., Qian, B., Wen, Z., Shah, T., Morgan, G., and

Ranjan, R. (2023). Explainable ai (xai): Core ideas,

techniques, and solutions. ACM Comput. Surv., 55(9).

Fasano, F., Addante, A. S., Valenzano, B., and Scannicchio,

G. (2021). Variables inﬂuencing per capita produc-

tion, separate collection, and costs of municipal solid

waste in the apulia region (italy): An experience of

deep learning. International Journal of Environmen-

tal Research and Public Health, 18(2):752.

Jevinger,

A., Zhao, C., Persson, J. A., and Davidsson,

P. (2024). Artiﬁcial intelligence for improving pub-

lic transport: a mapping study. Public Transport,

16(1):99–158.

Kalampokis, E., Karamanou, A., and Tarabanis, K. (2021).

Applying explainable artiﬁcial intelligence techniques

on linked open government data. In Scholl, H. J., Gil-

Garcia, J. R., Janssen, M., Kalampokis, E., Lindgren,

I., and Rodr

ıguez Bol

ıvar, M. P., editors, Electronic

Government, pages 247–258, Cham. Springer Inter-

national Publishing.

Karamanou, A., Kalampokis, E., and Tarabanis, K. (2022).

Linked open government data to predict and explain

house prices: the case of scottish statistics portal. Big

Data Research, 30:100355.

Lnenicka, M. and Nikiforova, A. (2021). Transparency-by-

design: What is the role of open data portals? Telem-

atics and Informatics, 61:101605.

Orusa, T., Viani, A., and Borgogno-Mondino, E. (2024).

Earth observation data and geospatial deep learning

ai to assign contributions to european municipalities

sen4mun: an empirical application in aosta valley (nw

italy). Land, 13(1):80.

Papadakis, T., Christou, I. T., Ipektsidis, C., Soldatos, J.,

and Amicone, A. (2024). Explainable and transparent

artiﬁcial intelligence for public policymaking. Data

& Policy, 6:e10.

Park, S. and Gil-Garcia, J. R. (2022). Open data innovation:

Visualizations and process redesign as a way to bridge

the transparency-accountability gap. Government In-

formation Quarterly, 39(1):101456.

Rashid, A. B. and Kausik, A. K. (2024). Ai revolutioniz-

ing industries worldwide: A comprehensive overview

of its diverse applications. Hybrid Advances, page

100277.

Roseck

y, M.,

Sompl

ak, R., Slav

ık, J., Kalina, J., Bulkov

G., and Bedn

r, J. (2021). Predictive modelling as a

tool for effective municipal waste management policy

at different territorial levels. Journal of Environmental

Management, 291:112584.

Wani, N. A., Kumar, R., Mamta, Bedi, J., and Rida, I.

(2024). Explainable ai-driven iomt fusion: Unrav-

elling techniques, opportunities, and challenges with

explainable ai in healthcare. Information Fusion,

110:102472.

An Explainable Model for Waste Cost Prediction: A Study on Linked Open Data in Italy

453