Explainable Recommendations of Drugs for Diabetic Patients

Priscila Valdiviezo-Diaz

Department of Computer Science, Universidad T

ecnica Particular de Loja, Loja, Ecuador

Keywords:

Diabetes, Drug, Collaborative Filtering, Explainable Recommendation, Recommender System.

Abstract:

Currently, recommender systems are widely used for different purposes, for example, to recommend resources,

products, and services. In the health domain, recommender systems are being used to recommender drugs,

treatments, food plans, and healthcare services in general. Collaborative ﬁltering is the most popular technique

in the recommender system area. This technique can be of two types: memory-based collaborative ﬁltering and

based-model collaborative ﬁltering. One of the problems of recommender systems is that most of them focus

on enhancing the precision of the recommendation and do not provide a justiﬁcation for the suggestions given

to the user. Therefore, it is important to provide explainable recommendations so that the user understands

why an item is recommended. To address this problem, in this paper the use of a Bayesian method for

explainable drug recommendations for diabetic patients is presented. Several experiments are carried out using

a dataset with information on diabetic patients with three collaborative ﬁltering approaches: the memory-based

approach IbCF, and two model-based approaches: item-based NBCF, and Hybrid NBCF. The experimental

results present good results for the Hybrid NBCF approach compared to the other approaches tested. Moreover,

it is observed a better quality of prediction and an increase in recommendation precision with Hybrid NBCF.

1 INTRODUCTION

The advancement of technology has allowed health

institutions to have a large amount of information

about patients, which can be analyzed and used to

help doctors prescribe treatment and medication prop-

erly. This information, often available in medical

databases, refers to laboratory test results, treatments,

diagnoses, and prescribed medications. According to

(Wang et al., 2022), it is scientiﬁcally important to use

drugs to improve their effectiveness in disease treat-

ment.

In this sense, recommender systems (RS) are been

widely applied to the health domain to support medi-

cal suggestions and provide personalized attention to

the patient (Tran et al., 2019). Recent research on

RS on the health has focused on disease prediction

and recommending the precautions (Rustam et al.,

2022), content recommendations to patients with dia-

betes (Nagaraj and Deepalakshmi, 2022), recommen-

dations for national ﬁtness items (Li and Yang, 2022),

and recommendations for healthcare services (Meng

et al., 2022). Nowadays, recommender systems de-

veloped within the health care setting, have been used

for disease management programs, for example, hy-

https://orcid.org/0000-0002-5216-8820

pertension (Sajde et al., 2022), and diabetes (Kamath

et al., 2022), providing a personalized user experi-

ence.

In the health domain, it is also important to have

explainable machine learning models that help health

professionals in decision-making, and internal ac-

tions, thus, by explaining the results of a prediction,

the trust of doctors is gained making it possible to ap-

ply the predictive model in practical situations (Yang,

2022). These models are being used in recommender

systems to provide justiﬁcation for the suggestions

that the system provides.

On the other hand, one of the diseases that has

attracted a lot of attention from health researchers

is diabetes. Diabetes is a chronic disease caused

by a lack of physical activity and unhealthy eating

habits (Bankhele et al., 2017). In this context, (Ali

et al., 2018) develop a recommender system to sug-

gest physical activity and diet plans to help patients

control this disease and avoid future complications.

These systems also help health professionals to

provide medical recommendations on treatments or

medications to prescribe to the patient. According

to (Calero Valdez et al., 2016), these systems are ex-

pected to minimize time and effort in the healthcare

decision-making process. Although some works re-

Valdiviezo-Diaz, P.

Explainable Recommendations of Drugs for Diabetic Patients.

DOI: 10.5220/0011617400003393

In Proceedings of the 15th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2023) - Volume 3, pages 73-80

ISBN: 978-989-758-623-1; ISSN: 2184-433X

 2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)

lated to recommender systems have been developed

in the health domain, most focus on the recommen-

dation of a speciﬁc item or service and enhancing the

recommendation’s precision. Still, they do not pro-

vide an explanation of why a particular drug is rec-

ommended.

Traditional recommendation systems are based on

machine learning techniques such as matrix factoriza-

tion (MF) (Azri et al., 2023), KNN (Nagaraj et al.,

2022), and neural networks (Chaithra et al., 2023).

However, models based on MF and neural networks

are difﬁcult to explain the recommendations. Accord-

ing to (Ammar and Shaban-Nejad, 2020), explain-

able machine learning models promote credibility and

trust in critical areas, such as medicine, by combin-

ing machine learning techniques that explicitly show

why a recommendation is made. Therefore, unlike the

works reviewed in the state-of-the-art, in this paper, in

addition to providing a list of medications for diabetic

patients, an explanation is provided for these recom-

mendations. Previous this, an analyzing the perfor-

mance of three collaborative ﬁltering recommenda-

tion methods which facilitate the explanation of rec-

ommendations is carried out, and then the approach

that better results present is selected.

Explaining the recommendations to the patient

can improve users’ trust and thus increase acceptance

of the system by health personnel. In accordance with

(Tran et al., 2021), trust is even more critical for RS

to convince patients to follow health-related recom-

mendations. This aspect can be enhanced by provid-

ing explanations for recommendations (Ammar and

Shaban-Nejad, 2020).

The rest of the paper is structured as follows: Sec-

tion II presents related work. Section III encloses the

context of the present work. Section IV includes the

material and method used for the experiments. Sec-

tion V shows the experimental results and the process

for the explanation of recommendations. Section VI

encloses the conclusions and future work.

2 RELATED WORK

In this section, studies focusing on the development

of recommender systems for diabetes, medicine rec-

ommendation, and works including explaining rec-

ommendations are presented.

2.1 Recommender Systems for Diabetes

Patients

Several studies on recommender systems have been

developed for the treatment and control of diabetes.

For example, (Zeng et al., 2017) use information

retrieval approaches in recommending diabetic pa-

tient education materials based on diabetic questions

posted on the TuDiabetes forum. In (Bankhele et al.,

2017) propose an android application based on a

user-based collaborative ﬁltering approach to suggest

probable medication, diet, and exercise to help people

manage their diabetes well. This application can also

remind users to carry out the recommendations which

are provided by the system. Authors in (Rehman

et al., 2017) present a cloud-based food recommen-

dation system, for dietary recommendations based on

users’ pathological reports. An ant colony algorithm

is used to generate an optimal food list and recom-

mends suitable foods according to the values of patho-

logical reports. Likewise, in (Bhat and Ansari, 2021)

a machine learning technique is used for diagnosis of

diabetes and recommend proper diet for diabetic pa-

tient.

(Nagaraj and Deepalakshmi, 2022) propose an in-

telligent fuzzy inference rule-based predictive dia-

betes diagnosis model, providing content recommen-

dations to patients with diabetes. The model pre-

dicts the risk of diabetes disease using fuzzy infer-

ence based on Mamdani’s technique, then the recom-

mendations for a normal life, nutrition, exercise, and

medications are given to patients.

In (Almulla, 2020) propose an expert system to

diagnoses diabetes and recommends the right med-

ication depending on the location where the patient

lives and on the symptoms of the patient and other ef-

fective factors. The system outputs a list of names of

locally available brand names of medications that suit

the diabetes type.

2.2 Medicine Recommender Systems

In recent years research related to medication recom-

mendations based on machine learning has been de-

veloped, for example, a medicine recommendation

model based on the incorporation of graphs to rec-

ommend appropriate drugs for patients is proposed in

(Wang et al., 2022). This model generates the rec-

ommended drug list by calculating the cosine similar-

ity between disease combination representations and

drug combination representations. A recommenda-

tion algorithm called LEAP is presented in (Zhang

et al., 2017), which uses records of current patient vis-

its and drug-drug interactions to predict a list of med-

ICAART 2023 - 15th International Conference on Agents and Artiﬁcial Intelligence

ications. This algorithm uses a recurrent decoder to

model label dependencies and content-based attention

is used to capture label instance mapping. (Wedagu

et al., 2020) propose a recommendation method called

DIMERS, which combines a prior medical knowl-

edge of doctors with bidirectional Long Short-Term

Memory (BiLSTM). In this study, authors use a

weighted block with prior medical knowledge to en-

hance the learning of deep neural networks.

A drug recommendation model based on mes-

sage propagation neural network is proposed in (Ren

et al., 2022), in this paper, the Drug-Drug Interac-

tion (DDI) knowledge is introduced into the model

to reduce the DDI rate of recommended results. A

two-stage personalized medication recommender sys-

tem is presented by (Bhoi et al., 2021). Authors use

various weights in the system to compute the contri-

butions from the information sources for the recom-

mended medications. The system models the drug in-

teraction from an external drug database and the drug

co-occurrence from the electronic health records as

graphs.

2.3 Explainable Recommender Systems

In the health domain, a low number of papers have

been found to focus on providing explainable rec-

ommendations, some of them are oriented to recom-

mend food and explain these recommendations, for

example, (Padhiar et al., 2021) include explanations

to users for food-related suggestions. These authors

model food recommendations, using concepts from

the explanation domain to create responses to user

questions about food recommendations. Likewise, in

(Pecune et al., 2022) a conversational system that rec-

ommends recipes aligned with its users’ eating habits

and current preferences is presented. This system is

also able to justify its recipes recommendation by ex-

plaining the trade-off between them.

Authors in (Zoppis et al., 2019) present a com-

putational model for promoting targeted communica-

tion and supplying social explainable recommenda-

tions, in order to support the formation of commu-

nities of patients and health services. In (Guti

errez

et al., 2022) present the design and implementation of

a recommender engine and a mobile application de-

signed to support call recommendations and explain

these recommendations.

In the work of (Cai et al., 2022), a model of ex-

plainable recommendation on account of knowledge

graph as well as many-objective evolutionary algo-

rithm is proposed, which combines recommendation

and explanation. Likewise, (Chicaiza and Valdiviezo-

diaz, 2022) present a research related to explain-

able recommender systems. In this research, the au-

thors describe two scenarios based on the Tripadvisor

dataset to generate restaurant explainable recommen-

dations. Explainable recommendations are evaluated

using Fidelity and Transparency metrics.

3 CONTEXT

A previous work on drug recommendations for dia-

betes patients was presented without considering ex-

plaining the recommendation and using other predic-

tion techniques. This previous work is based on col-

laborative ﬁltering (CF) and clustering techniques for

recommending drugs to diabetes patients (Morales

et al., 2022). This system realizes suggestions accord-

ing to drug information and the characteristics of pa-

tients. The clustering technique is applied to group

patients with similar characteristics, and the collabo-

rative ﬁltering technique is applied to represent the

patient’s explicit data, then based on the group to

which the patient belongs, the recommendation is

made considering the drugs with the highest predic-

tion value.

Currently, Diabetes is a chronic metabolic disease

that generates a great impact on the world population.

According to (Saeedi et al., 2019) it disease is among

the top 10 causes of death in adults. Diabetes can

be treated through physical activity, healthy eating,

medication, and regular checkups to prevent compli-

cations.

In (Association, 2020) different types of medica-

tions recommended by the American Diabetes Asso-

ciation (ADA) related to or used in the treatment of

this condition are presented. It is possible to deter-

mine that there are many different types of drugs that

can work to lower your blood sugar. Sometimes one

medication is enough, but the doctor may prescribe a

combination of medicines in other cases.

Authors in (Al-Soﬁani et al., 2021) propose a

medication algorithm scheme for the treatment of

people with Diabetes. Authors put special empha-

sis on medication cost and medication adherence as

determining factors in the choice of diabetes medica-

tions recommended.

The aim is to use the information on essential

medicines and doses prescribed for diabetes patients

to make drug explainable recommendations for treat-

ing this disease. Therefore, in the present work, a

step forward from the drug recommendation is taken,

considering the justify the recommendation given to

the patient based on the diabetic patient’s explicit

data (patient-dose-drug), and using probabilistic algo-

rithms that allow understanding of the recommenda-

Explainable Recommendations of Drugs for Diabetic Patients

tion. As a result, three collaborative ﬁltering algo-

rithms were tested using a dataset with diabetic pa-

tient information.

In the work presented by (Valdiviezo-Diaz et al.,

2019) a bayesian hybrid approach facilitates under-

standing and explaining the recommendation of the

user. This CF approach called Naive Bayes Collabo-

rative Filtering (NBCF) recommends items by using

similar users’ and items’ information, respectively.

This novelty approach has been considered to ex-

plain recommendations taking into consideration the

information of patients and the drugs speciﬁed in the

dataset used, and instead of the rating, the dose of

the drugs is considered. Moreover, this approach has

been selected because it provides successful results in

the quality of recommendations.

4 MATERIAL AND METHOD

Many machine learning techniques have been de-

signed for diabetes diagnosis, the prediction of this

disease, and providing useful analysis of medical

data. In our work, a probabilistic machine learning

model for explainable drug recommendation for dia-

betic patients is used.

For the recommendation, the collaborative ﬁlter-

ing recommendation approach will be use, since, ac-

cording to (Wang et al., 2020) this approch is one of

the most applied methods in the health recommen-

dation systems. The collaborative ﬁltering method

recommends to the active user items that other users

with similar preferences have liked in the past (Ricci

et al., 2015). CF can be of two types: memory-

based CF and model-based CF (Yang et al., 2016).

In the memory-based CF method, the RS uses the rat-

ings to ﬁnd neighbors for the target user or item and

computes the predicted value for the unknown rating.

Model-based CF uses a model to predict the value for

the unknown ratings based on the rating matrix.

In this manuscript, a memory-based CF method

and two model-based CF methods that allow recom-

mendation explanation are tested.

For a better demonstration of the explanation of

the medications, a recommendation scenario centered

on the patient is presented, to whom some medica-

tions are suggested by the system with their respective

justiﬁcation.

For the evaluation, the most common metrics will

be used to evaluate the performance of recommen-

dation systems. The prediction and recommendation

quality are computed using an existing dataset with

information on diabetic patients.

4.1 Dataset

For the experiments, a dataset with diabetic pa-

tient records available in the UCI Machine Learning

Repository is used, which refers to 100,000 observa-

tions and 50 features representing patient and hospital

outcomes (Dua and Graff, 2017). This dataset con-

tains information related to the personal data of the

patients, information about admission, procedures,

medications, and diagnostics results (Strack et al.,

2014).

The dataset was pre-processed to construct the

user-item rating matrix necessary in collaborative ﬁl-

tering. Patients who have been administered at least

two medications were selected, leaving a total of

5,148 unique patients. Likewise, a selection of drugs

was made considering those that have been prescribed

to at least 50% of the selected patients, as a result,

there is a total of 10 drugs. Then, to represent the

matrix of collaborative ﬁltering (patient-drug-dose),

the information on the patient’s dose for each drug is

considered, that is, if the dose remains stable, or if the

dose is increased.

Table 1 shows a summary of the data to be consid-

ered for the experimentation. Each drug is on a scale

from 1 to 2 (1: indicates whether the dose is main-

tained for the patient; 2: if the dose is increased).

Figure 1 shows the drugs selected for the experi-

ments and the percentage of patients using the drug.

From ﬁgure 1, Insulin is the drug that has been

prescribed the most to patients for the treatment of

diabetes, on the contrary, Glyburide-metformin and

Nateglinide are the drugs that have been prescribed

the least.

Table 1: Dataset Information.

Value Description

#Patients 5,148 Total patients

#Drugs 10 Total Drugs

#Ratings (dose) 46,593 1-2 values

4.2 Evaluation Metrics

To assess the accuracy of the prediction of algorithms,

Mean Absolute Error (MAE) (Wang and Lu, 2018) is

used. Also, Precision and Recall metrics are used to

evaluate the quality of the recommendations. Preci-

sion represents the percentage of recommended items

being relevant, and Recall represents the percentage

of relevant items being recommended (Valdiviezo-

ıaz and Bobadilla, 2019).

In addition to evaluating the quality of the predic-

tion and the recommendation, this paper presents the

ICAART 2023 - 15th International Conference on Agents and Artiﬁcial Intelligence

!"#$%&

'(#)%&

'*#$%&

'+#!%&

,#*%&

)#"%&

!#'%&

'#!%&

+#,%& +#,%&

-./012.&

34567892.&

:12;2<2=4&

:1>?082=4&

@27A125B<7.4&

C7/2A125B<7.4&

:1294;282=4&

C4;BA12.2=4&

:1>?082=4D94567892.&

EB54A12.2=4&

!"#$%&'$()*

Figure 1: Selected Drugs from the diabetes patients dataset.

evaluation of the explanations given by the algorithm.

To evaluate the quality of the explainable recommen-

dations Fidelity metric is used, which measures the

percentage of explainable items in the recommended

items (Peake and Wang, 2018).

5 RESULTS

This section presents the results of the experiments

with the dataset used, and the detail on how to explain

the recommendations with the selected method.

The cross-validation method is used to evaluate

the performance of the algorithms, where the dataset

is split into 80% training set and 20% testing set.

5.1 Performance Comparison

In order to select an algorithm for the explanation

of recommendations, we compare the performance

de three collaborative ﬁltering methods for recom-

mender systems, for example, model-based CF meth-

ods: Item-based NBCF and Hybrid NBCF, and the

traditional CF method based on memory IbCF. These

methods were selected for experimentation because

currently, model-based approaches are achieving bet-

ter results in accuracy and performance (Valdiviezo-

Diaz et al., 2019), and also the Bayesian models-

based methods provide a probabilistic interpretation

of their results, which facilitates explaining the deci-

sion process (Cheng et al., 2017). On the other hand,

memory-based approaches are that they are simple to

implement and the resulting recommendations are of-

ten easy to explain (Charu, 2016).

A more detailed description of the methods se-

lected is presented, join with the relevant hyper-

parameters set for the experiments.

• Item-based NBCF recommends items to the user

according to the ratings received by each item

(Valdiviezo-Diaz et al., 2019). The hyperparame-

ter set for this method is α = 0.01.

• Hybrid NBCF combines user-based NBCF and

item-based NBCF approaches to complement

each other and improve the accuracy of the model

(Valdiviezo-Diaz et al., 2019). The hyperparam-

eter set for this method is α = 0.01. This ap-

proach ﬁrst computes the prior distributions and

the likelihood for user-based NBCF and item-

based NBCF approaches, then both approaches

are combined using a weighted product.

• Item-based CF (IbCF) recommends similar items

to the item the active user has already preferred

in the past (Kant and Mahara, 2018). The cosine

correlation coefﬁcient is used as a similarity mea-

sure.

The hyperparameter values were selected in order

to maximize the accuracy of algorithms for quality

measures used.

In NBCF, the explanation is realized considering

the evidence set of the bayesian method. On the other

hand, the IbCF algorithm can explain the recommen-

dations provide, considering the similarity between

the items.

Table 2 shows the results of the comparative anal-

ysis of the CF algorithms based on the performance

evaluation metrics on the dataset used.

From the experimental results, it is observed that:

NBCF methods present a better performance in con-

trast to IbCF. The results show that Hybrid NBCF

outperforms all the other methods in terms of MAE.

A smaller value of this metric means better perfor-

mance. Moreover, it indicates that the predicted and

actual values are closer. Therefore, we can conclude

that the NBCF model is making a good prediction of

the drug dose.

Likewise, the results show more accurate values

concerning Precision for the probabilistic methods

(NBCF) in comparison to IbCF. From the table 2

we can see that 74% of the recommended drugs are

relevant or adequate for the patient. However, Re-

call results show that IbCF is better than the other

two methods tested. Therefore, analyzing the per-

formance of the three CF methods, Hybrid NBCF is

selected for the explanation of recommendations be-

cause it presents better performance in most of the

Table 2: Results of the metrics of each recommendation

algorithm for the dataset used.

Algorithm MAE Precision Recall

Item-based NBCF 0.356 0.695 0.594

Hybrid NBCF 0.332 0.744 0.616

IbCF 0.611 0.530 0.798

Explainable Recommendations of Drugs for Diabetic Patients

metrics calculated, for example: in MAE and Preci-

sion.

5.2 Explainable Recommendations for

Diabetic Patients

The selected algorithm Hybrid NBCF allows explain-

ing the predictions. For the explanation, what is men-

tioned in (Valdiviezo-Diaz et al., 2019) is considered,

which indicates that to explain a recommendation to

the user u is necessary to consider the case in which

the system has recommended the item with an esti-

mated rating, in our case, would be that the system has

recommended a drug with an estimated dose value.

Based on the algorithm for the Hybrid NBCF ap-

proach, we have the following recommendation for a

patient:

The case in which the system has recommended

the drug i with an estimated dose ˆr = 2 is considered.

As explained in (Valdiviezo-Diaz et al., 2019) is nec-

essary to obtain the P and Q evidences corresponding

to the user-based NBCF and item-based NBCF ap-

proaches, respectively. So, ﬁrstly, all drugs that have

been prescribed to patient u according to their likeli-

hood within the drug i are sorted from highest to low-

est. Secondly, the items from the list whose dose has

been increased for the patient are extracted, and add

them to the set of P evidences, in this case: Insulin.

Next, all patients who have been prescribed the drug

i according to their likelihood within the patient u are

sorted from highest to lowest. Then the patients of the

list whose dose of medication i has been increased are

extracted and added to the list of Q evidences, in our

case, patients: 2844, 3019, 3286, 3751. Finally, both

sets of evidence are combined by adjusting the P and

Q values according to the needs.

Figure 2 presents the explain the recommendation

given to a patient.

From ﬁgure 2 can be observed that in addition to

presenting the recommendation to the user, an expla-

nation of the recommendation is shown. It is hoped

that this explanation can help the user understand why

the drug is recommended and give the user greater

conﬁdence in the system.

"We recommend Metformin because you were

previously prescribed with the drug Insulin,

and the medication (Metformin) was prescribed

to users: 2844, 3019, 3286, 3751 who share

clinical manifestations with you”.

!"#$%&'()

Recommendation

Explanation

Figure 2: Recommendation explanation example.

5.3 Evaluation of the Explainable

Recommendations

(Zhang and Chen, 2018) introduce approaches to

evaluate recommendation explanations. The ﬁrst

approach evaluates the percentage of recommended

items that can be explained by the explainable rec-

ommendation model, regardless of the quality of the

explanations; and the second evaluates the quality of

the explanations exactly. This paper applies the ﬁrst

approach to evaluate the explication of the Hybrid

NBCF recommendation algorithm used to explain-

able drug recommendations.

In this section, we present the evaluation of the

recommendations using the Fidelity metric. In our

case, the recommended items will be the items whose

estimated dose is 2, that is, those items whose dose

has been increased. The explainable items will be

those items that are recommended but that can be ex-

plained because there is at least one element in the set

of P and Q evidences.

Therefore, applying the equation deﬁned in

(Peake and Wang, 2018), the ﬁdelity obtained is 0.66.

This means that evaluating the quality of the explana-

tions of the NBCF algorithm will depend on the exis-

tence of P and Q evidences. The more items that can

be explained, the higher the ﬁdelity value.

6 CONCLUSIONS

In the health domain, explainability is essential to

gain the trust of healthcare professionals and patients

and to enhance the transparency of the recommender

system.

In this paper, the use of a probabilistic approach

to explain recommendations based on the doses of

drugs prescribed to diabetic patients was presented.

Because this approach is probabilistic, we think that

it makes the explanations easily understandable to

users. We have also established how would be the

explanation of the recommendation made to a user

within the system with the Hybrid NBCF approach,

and how to evaluate the explanations determining

the faithfulness of the explainable recommendation

model.

We focused our research on testing CF meth-

ods which allow the explanation of recommendations.

From tested methods, NBCF shows betters results in

drug dose prediction accuracy. Moreover, the results

of experiments conducted on a real dataset of dia-

betic patients verify the good recommendation per-

formance and explanatory ability of Hybrid NBCF.

ICAART 2023 - 15th International Conference on Agents and Artiﬁcial Intelligence

Future work includes: a) testing other collabo-

rative ﬁltering methods for explaining recommenda-

tions using the same dataset, and b) evaluating the

transparency of the recommender system, providing

to the patients an understanding of how the system

formulated the recommendation.

REFERENCES

Al-Soﬁani, M., Batais, M., Aldhaﬁri, E., and Alzaid, A.

(2021). Pharmacological treatment of type 2 diabetes

in saudi arabia: A consensus statement from the saudi

society of endocrinology and metabolism (ssem). Di-

abetes & Metabolic Syndrome: Clinical Research &

Reviews, 15.

Ali, S. I., Amin, M. B., Kim, S., and Lee, S. (2018). A

hybrid framework for a comprehensive physical ac-

tivity and diet recommendation system. In Mokhtari,

M., Abdulrazak, B., and Aloulou, H., editors, Smart

Homes and Health Telematics, Designing a Better Fu-

ture: Urban Assisted Living, pages 101–109, Cham.

Springer International Publishing.

Almulla, M. (2020). Location-based expert system for

diabetes diagnosis and medication recommendation.

Kuwait Journal of Science, 48.

Ammar, N. and Shaban-Nejad, A. (2020). Explainable ar-

tiﬁcial intelligence recommendation system by lever-

aging the semantics of adverse childhood experiences:

Proof-of-concept prototype development. Journal of

Medical Internet Research, 8:e18752.

Association, A. D. (2020). Erratum: Addendum. 9. phar-

macologic approaches to glycemic treatment: Stan-

dards of medical care in diabetes-2020. diabetes care

2020;43(suppl. 1):s98-s110 (diabetes care (2020) 43

suppl 1 (s98-s110)). Diabetes care, 43(8):1979. Cited

By :6.

Azri, A., Haddi, A., and Allali, H. (2023). IU-AutoSVD++:

An Item–User Features-Based Recommender System

Using Contractive Autoencoder and Matrix Factor-

ization, volume 448 of Lecture Notes in Networks and

Systems.

Bankhele, S., Mhaske, A., Bhat, S., and V., S. (2017). A

diabetic healthcare recommendation system. Interna-

tional Journal of Computer Applications, 167:14–18.

Bhat, S. S. and Ansari, G. A. (2021). Predictions of di-

abetes and diet recommendation system for diabetic

patients using machine learning techniques. In 2021

2nd International Conference for Emerging Technol-

ogy (INCET), pages 1–5.

Bhoi, S., Lee, M. L., Hsu, W., Fang, H. S. A., and Tan,

N. C. (2021). Personalizing medication recommenda-

tion with a graph-based approach. ACM Trans. Inf.

Syst., 40(3).

Cai, X., Xie, L., Tian, R., and Cui, Z. (2022). Explicable

recommendation based on knowledge graph. Expert

Systems with Applications, 200:117035.

Calero Valdez, A., Zieﬂe, M., Verbert, K., Felfernig, A.,

and Holzinger, A. (2016). Recommender Systems for

Health Informatics: State-of-the-Art and Future Per-

spectives, pages 391–414. Springer International Pub-

lishing, Cham.

Chaithra, Lingaraju, G. M., and Jagannatha, S. (2023). Web

page recommendation using distributional recurrent

neural network. Computer Systems Science and En-

gineering, 45(1):803–817.

Charu, A. (2016). The Textbook. In Recommender Systems,

pages XXI, 498. Springer Cham, ﬁrst edit edition.

Cheng, S., Yang, H., and Shafto, P. (2017). Explainable

Artiﬁcial Intelligence via Bayesian Teaching. In 31st

Conference on Neural Information Processing Sys-

tems (NIPS 2017), Long Beach, CA, USA.

Chicaiza, J. and Valdiviezo-diaz, P. (2022). Explainable

Recommender Systems : From Theory to Practice. In

6th World Conference on Smart Trends in Systems, Se-

curity and Sustainabilit, pages 1–10.

Dua, D. and Graff, C. (2017). UCI machine learning repos-

itory.

Guti

errez, F., Htun, N. N., Vanden Abeele, V., De Croon,

R., and Verbert, K. (2022). Explaining call recom-

mendations in nursing homes: A user-centered design

approach for interacting with knowledge-based health

decision support systems. In 27th International Con-

ference on Intelligent User Interfaces, IUI ’22, page

162–172, New York, NY, USA. Association for Com-

puting Machinery.

Kamath, S. S., Kappaganthu, K., Painter, S. L., and Madan,

A. (2022). Improving outcomes through personal-

ized recommendations in a remote diabetes monitor-

ing program: Observational study. JMIR Formative

Research, 6.

Kant, S. and Mahara, T. (2018). Merging user and item

based collaborative ﬁltering to alleviate data sparsity.

International Journal of Systems Assurance Engineer-

ing and Management, 9(1):173–179.

Li, L. and Yang, A. (2022). A Recommendation Method

of National Fitness Items Based on Neural Network

Algorithm. Scientiﬁc Programming, 2022:1544767.

Meng, S., Fan, S., Li, Q., Wang, X., Zhang, J., Xu, X.,

Qi, L., and Bhuiyan, M. Z. A. (2022). Privacy-aware

factorization-based hybrid recommendation method

for healthcare services. IEEE Transactions on Indus-

trial Informatics, 18(8):5637–5647.

Morales, L. F. G., Valdiviezo-Diaz, P., Re

A¡tegui, R., and

Barba-Guaman, L. (2022). Drug recommendation

system for diabetes using a collaborative ﬁltering and

clustering approach: Development and performance

evaluation. Journal of Medical Internet Research,

24(7). Cited by: 0; All Open Access, Gold Open Ac-

cess, Green Open Access.

Nagaraj, P. and Deepalakshmi, P. (2022). An intelli-

gent fuzzy inference rule-based expert recommenda-

tion system for predictive diabetes diagnosis. Inter-

national Journal of Imaging Systems and Technology,

32(4):1373–1396.

Nagaraj, P., Saiteja, K., Ram, K. K., Kanta, K. M., Aditya,

S. K., and Muneeswaran, V. (2022). University recom-

mender system based on student proﬁle using feature

Explainable Recommendations of Drugs for Diabetic Patients

weighted algorithm and knn. In International Con-

ference on Sustainable Computing and Data Commu-

nication Systems, ICSCDS 2022 - Proceedings, pages

479–484. Cited By :2.

Padhiar, I., Seneviratne, O., Chari, S., Gruen, D., and

McGuinness, D. L. (2021). Semantic modeling for

food recommendation explanations.

Peake, G. and Wang, J. (2018). Explanation mining: Post

hoc interpretability of latent factor models for rec-

ommendation systems. In Proceedings of the 24th

ACM SIGKDD International Conference on Knowl-

edge Discovery & Data Mining, KDD ’18, page

2060–2069, New York, NY, USA. Association for

Computing Machinery.

Pecune, F., Callebert, L., and Marsella, S. (2022). Design-

ing persuasive food conversational recommender sys-

tems with nudging and socially-aware conversational

strategies. Frontiers in Robotics and AI, page 390.

Rehman, F., Khalid, O., Haq, N., Khan, A. u. R., Bilal, K.,

and Madani, S. (2017). Diet-right: A smart food rec-

ommendation system. KSII Transactions on Internet

and Information Systems, 11:2910–2925.

Ren, Y., Shi, Y., Zhang, K., Wang, X., Chen, Z., and

Li, H. (2022). A drug recommendation model based

on message propagation and ddi gating mechanism.

IEEE Journal of Biomedical and Health Informatics,

26(7):3478–3485.

Ricci, F., Rokach, L., and Shapira, B. (2015). Recommender

Systems Handbook, volume 32.

Rustam, F., Imtiaz, Z., Mehmood, A., Rupapara, V., Choi,

G. S., Din, S., and Ashraf, I. (2022). Automated dis-

ease diagnosis and precaution recommender system

using supervised machine learning. Multimedia Tools

and Applications, 81(22):31929–31952.

Saeedi, P., Petersohn, I., Salpea, P., Malanda, B., Karu-

ranga, S., Unwin, N., Colagiuri, S., Guariguata, L.,

Motala, A. A., Ogurtsova, K., Shaw, J. E., Bright,

D., Williams, R., and behalf of the IDF Diabetes At-

las Committee, O. (2019). Global and regional di-

abetes prevalence estimates for 2019 and projections

for 2030 and 2045: Results from the international di-

abetes federation diabetes atlas, 9th edition. Diabetes

research and clinical practice, 157. Cited By :3061.

Sajde, M., Malek, H., and Mohsenzadeh, M. (2022). Re-

comed: A knowledge-aware recommender system for

hypertension medications. Informatics in Medicine

Unlocked, 30:100950.

Strack, B., DeShazo, J. P., Gennings, C., Olmo, J. L., Ven-

tura, S., Cios, K. J., and Clore, J. N. (2014). Impact of

HbA1c Measurement on Hospital Readmission Rates:

Analysis of 70,000 Clinical Database Patient Records.

BioMed Research International, 2014:781670.

Tran, T. N. T., Atas, M., Felfernig, A., Le, V. M., Samer,

R., and Stettinger, M. (2019). Towards social choice-

based explanations in group recommender systems.

In Proceedings of the 27th ACM Conference on User

Modeling, Adaptation and Personalization, UMAP

’19, page 13–21, New York, NY, USA. Association

for Computing Machinery.

Tran, T. N. T., Felfernig, A., Trattner, C., and Holzinger,

A. (2021). Recommender systems in the healthcare

domain: state-of-the-art and research issues. Journal

of Intelligent Information Systems, 57:1–31.

Valdiviezo-D

ıaz, P. and Bobadilla, J. (2019). A hybrid ap-

proach of recommendation via extended matrix based

on collaborative ﬁltering with demographics informa-

tion, volume 895 of Communications in Computer

and Information Science. Cited By :5.

Valdiviezo-Diaz, P., Ortega, F., Cobos, E., and Lara-

Cabrera, R. (2019). A collaborative ﬁltering ap-

proach based on na

ıve bayes classiﬁer. IEEE Access,

7:108581–108592. Cited By :27.

Wang, N., Cai, X., Yang, L., and Mei, X. (2022).

Safe medicine recommendation via star interactive

enhanced-based transformer model. Computers in Bi-

ology and Medicine, 141:105159.

Wang, W. and Lu, Y. (2018). Analysis of the mean absolute

error (MAE) and the root mean square error (RMSE)

in assessing rounding model. IOP Conference Series:

Materials Science and Engineering, 324:012049.

Wang, Z., Huang, H., Cui, L., Chen, J., An, J., Duan, H.,

Ge, H., and Deng, N. (2020). Using natural language

processing techniques to provide personalized educa-

tional materials for chronic disease patients in china:

Development and assessment of a knowledge-based

health recommender system. JMIR Medical Informat-

ics, 8(4). Cited By :8.

Wedagu, M. A., Chen, D., Hussain, M. A. I., Gebremeskel,

T., Orlando, M. T., and Manzoor, A. (2020). Medicine

recommendation system for diabetes using prior med-

ical knowledge. In Proceedings of the 2020 4th In-

ternational Conference on Vision, Image and Signal

Processing, ICVISP 2020, New York, NY, USA. As-

sociation for Computing Machinery.

Yang, C. C. (2022). Explainable Artiﬁcial Intelligence for

Predictive Modeling in Healthcare. Journal of Health-

care Informatics Research, 6(2):228–239.

Yang, Z., Wu, B., Zheng, K., Wang, X., and Lei, L.

(2016). A survey of collaborative ﬁltering-based rec-

ommender systems for mobile internet applications.

IEEE Access, 4:3273–3287.

Zeng, Y., Liu, X., Wang, Y., Shen, F., Liu, S., Rastegar-

Mojarad, M., Wang, L., and Liu, H. (2017). Rec-

ommending education materials for diabetic questions

using information retrieval approaches. Journal of

Medical Internet Research, 19:e342.

Zhang, Y., Chen, R., Tang, J., Stewart, W. F., and Sun,

J. (2017). Leap: Learning to prescribe effective and

safe treatment combinations for multimorbidity. In

Proceedings of the 23rd ACM SIGKDD International

Conference on Knowledge Discovery and Data Min-

ing, KDD ’17, page 1315–1324, New York, NY, USA.

Association for Computing Machinery.

Zhang, Y. and Chen, X. (2018). Explainable recommenda-

tion: A survey and new perspectives. Found. Trends

Inf. Retr., 14:1–101.

Zoppis, I., Manzoni, S., and Mauri, G. (2019). A com-

putational model for promoting targeted communica-

tion and supplying social explainable recommenda-

tions. In 2019 IEEE 32nd International Symposium

on Computer-Based Medical Systems (CBMS), pages

429–434.

ICAART 2023 - 15th International Conference on Agents and Artiﬁcial Intelligence