2022 International Mobile and Embedded Technology Conference (MECON)

Disease Detection Using RASA Chatbot

Vaishali Gupta Advitiya Sood Dr. Tanya Singh
Department of Computer Science & Department of Computer Science & Department of Computer Science &
Engineering, ASET, Amity University, Engineering, ASET, Amity University, Engineering, ASET, Amity University,
Noida, UP, India. Noida, UP, India. Noida, UP, India. tsingh2@[Link]
vaishaligupta2000@[Link] advitiyasood@[Link]
2022 International Mobile and Embedded Technology Conference (MECON) | 978-1-6654-2020-4/22/$31.00 ©2022 IEEE | DOI: 10.1109/MECON53876.2022.9752338

Abstract - Chatbots, or conversational AI(Artificial early detection of serious diseases is a great challenge. Going
Intelligence) Interfaces, provide individuals a new way to to the hospital is a time consuming process and is not easily
interact with computer systems. Chatbots allow users to have affordable.
conversations with the system by asking questions in the way
that they would with another human beings. The current The medical chatbot overcomes all of these issues; the
adoption rate of chatbots on computer chat platforms is very main benefit of a chatbot is that it is free, easily available and
high. Such robots use artificial intelligence to understand convenient to use, meaning that a user can use it at at any time.
human input and respond accordingly. The core technology for As a result, it is a significant invention that primarily saves
the rise of chatbots is “Natural language processing” (NLP). The time spent consulting specialised doctors. It also responsible
recent advancements in NLP have allowed chatbots to be more to provides people with more information about their health
receptive than ever. Today, humans can interact with the status and preventative steps to help them stay healthy
chatbot systems anytime, anywhere. Chatbots can perform indefinitely.
predictive tasks (especially in the medical field), which is now
possible with advances in artificial intelligence and data mining Various data mining algorithms are available, and
technology. Healthcare, agriculture and education are when compared using medical datasets for performance and
important areas that need the most attention. In today's world, accuracy, studies showed that The Decision Tree, Nave
with the change in lifestyle and the current pandemic, illnesses Bayes, Neural Networks, KNN, SVM algorithms had the
have increased in the general population. As a result, the need ability to explore hidden pattern within the data. As abundance
for hospitals and doctors have increased substantially. Patients of medical data and information rich records are available for
have to spend their time waiting to be taken care of by the these algorithms to feed on a significant impact can be made
doctors. Also, doctors have an immense amount of workload in this industry.
with the amount of visits they have. Thus, the future of
healthcare depends on the ability of care providers to perform II. LITERATURE SURVEY
accurate remote diagnosis. This can be done by collecting data
remotely, and by using artificial intelligence to analyse data to
Chatbots are Computer programs equipped for doing
improve business and health outcomes. In this paper we regular discussions. Artificial intelligence is changing
analysed datasets to accurately detect diseases with organizations and chatbots controlled by AI are turning into
classification techniques such as SVM Classifier and Naïve an attainable client assistance channel and diminishes labor.
Bayes Classifier. Some utilization refined NLU frameworks, yet numerous less
complex ones examine for keywords inside the given
The tools used are RASA, machine learning classification information, then, at that point, pull an answer with the most
algorithms, data extraction etc. coordinating with catchphrases/designs from a data set.

Keywords: Artificial Intelligence, accurate remote diagnosis, A paper by Sherwin Fernandes [Link] [1] on doctor chatbot
RASA, Natural language processing. used classification models like SVM and Naïve Bayers, KNN
and Decision Tree an compared their accuracy on heart
I. INRODUCTION disease dataset and found out SVM had the highest accuracy.
A paper by Aishwarya Kedar [Link] [2] used PRISMA checklist
A healthy body is one of the greatest gifts we can have. methodology as the model of systematic review. They chose
Every one of us aspires to have a healthy physique and a better this method because PRISMA is the recognized standard for
quality of life. But, humans have been working so hard and reporting evidence in systematic reviews and meta-
constantly over the past few years that they have neglected to analyses.[3] A Chatbot for Psychiatric Counselling in Mental
prioritise their health on a daily basis. In the long run, this Healthcare used NLU, Word Embedding Algorithms,
issue puts people's quality of life in jeopardy. We can now Framework for Emotional Recognition. In this multiple-
give health care services to individuals at their convenience at modal approach is used for defining an Emotional Expressions
cheap prices all thanks to Artificial Intelligence. to Categorize and Collect Training Data for Emotion
A chatbot is an artificial intelligence-based software tool Recognition, Emotion Recognition, and Inference [4].
that may mimic a discussion (or a chat) with a user in human MedBot used Pattern matching algorithm, NLU, Dialogflow
language via messaging apps, websites etc. A chatbot is API and DoctorMe App’s datasets. Using provided API this
frequently regarded as one of the most advanced, effective chatbot was easily implemented in Facebook, Hangout etc.[5]
and promising forms of human-machine interaction. The heart In a self-diagnosing healthcare chatbot by S. Anil Kumar
and soul of AI-powered chatbots is natural NLP. The [Link] [6] A linear design was used that went from symptom
advanced NLP algorithms can understand, deduce, and extraction to symptom mapping, identifying the
identify the expression of incoming text by the client and can corresponding symptom, determining whether it was a major
respond accordingly. or minor disease, referring the patient to a doctor, extracting
People avoid going to the hospital for minor issues. In the the doctor's information from the database, and identifying the
future, these little issues will turn into serious diseases. The user using the login information from the database. With no

978-1-6654-2020-4/22/$31.00 ©2022 IEEE 94

Authorized licensed use limited to: Universitas Indonesia. Downloaded on November 19,2024 at 14:25:47 UTC from IEEE Xplore. Restrictions apply.
 2022 International Mobile and Embedded Technology Conference (MECON)

need to wait for a medical expert, travel, or even lose business shown on web app [13]. One significant undertaking done by
days, online doctor consultation overcomes geographic NLP is tokenization. The key usefulness of tokenization is to
barriers and provides professional understanding for the change over the text got from the client to tokens.
patient with their problem. . There are some ways to achieve Tokenization happens in such a manner that various words
weight loss success: increasing exercise, reducing food intake, will be changed over to various tokens.[14]. It proposes the
self-monitoring of diet, exercise as well as weight, self- possibility of chatbot as a virtual collaborator or as a shrewd
regulation.[7] specialist that can finish undertakings like giving legitimate
reactions to inquiries from clients, controlling gadgets, giving
In the paper by Lekha Athota [Link] [8] the quire of the user courses during driving and so forth.[16]
is answered using an expert system if the answer is not
available in the database. The space specialists additionally In M. Mandzuka et al. [17], the creators exhibited four
should enlist themselves by giving different subtleties. The state of the art assessment measurements prior to applying the
information of the chatbot put away in the data set as example most generally utilized text based arrangement methods like
format. Here SQL is used for dealing with the dataset. Neural Network, Random Forests, Bayesian Network, and
SVM. As per their exploration concentrate on SVM works
Here the chatbot is made for the client assistance that best much of the time for include determination. The
capacities as general wellbeing administration. The individuals from this review [18] utilized examination records
application utilizes Ngram, TF-IDF and cosine closeness. The in the text arrangement to assess the reason for death. They
information base is made for putting away the Q & A. The incorporated toxicology results from seven particular reasons
application unmistakably shows extricated the catchphrase for death, ran preprocessing, dimensionality decrease, and
from the inquiry and by utilizing unigram, bigram, and trigram component designing on them, and afterward made a bit work
which helps in quick replying.[9] with 20 highlights for each audit.
According to Du Preez [Link] [10] in which a voice A paper by Vikas Kamra [Link] [19] proposed a model that
acknowledgment talk bot is created. The inquiries posed to the accepting client's manifestations as information as a sound
bot are not perceived are further handled utilizing the expert-
record and anticipated the client's probability of creating
system of third parties. The webbots are made as web- mental problems dependent on that data. Not really settled
companions dependent on text, a client performer. In the event whether an individual has OCD, BPD, or Schizophrenia, just
that the program isn't just text-based, yet in addition voice- as its exactness.
based prepared, they focused on the improved framework
here. Here, a two-section interaction of catching and The proposed thought of the paper "An original
examining an information signal is needed for voice methodology for clinical help utilizing prepared chatbot" by
acknowledgment. Acknowledgment of information from the Divya Madhu [Link] [20] is to design a model using modernized
server reaction and handling of data. The server utilized here thinking eat urges the customer to see the ensured therapy for
is a discovery approach in view of SOAP. Utilizing a specialist affliction. There are a huge load of meds available for a
framework makes it conceivable to work on limitless and particular sickness and it's not possible for anyone to
independent knowledge. unequivocally suggest the fitting treatment and which is the
best partner for that disease. In this proposed, artificial mental
Mrs. Asha Rani [Link] [11] used a approach in which the ability accepts a critical part by giving a record of available
User will send queries and chatbot will give its reaction with medications
proper message. For the smooth exceeding everyone's
expectations customized predefined way with some III. PROPOSED METHODOLOGY
conceivable inquiry and answers that can be posed by the
client. This text/visit handling is finished utilizing Natural The purpose of this chatbot is to maintain human-like
Language Processing. The possibility of the paper is a self- interaction with the user along with predicting the correct
analysis clinical chatbot the utilization of computerized disease according to the user’s symptoms. NLP techniques
reasoning through [12] is to foster a machine the utilization of along with machine learning algorithms allow us to create
AI so one can assist clients to avoid meeting with a doctor. It such a conversational AI. While our main focus is to give a
is made to distinguish the disease of a client and convey prediction about the disease the user might have based on
significant subtleties of a sickness. their symptoms, creating an interactive medical bot which can
hold a conversation with the user and gives appropriate
The clinical consideration chatbots working depends upon responses and accurate information is also a feature of a good
Herbal language Processing that works with clients to set up and reliable chatbot. For all of this we need to have the correct
their bother roughly the wellness. inquiry is dispatched to algorithms as well as a good data set.
chatbot and the buyer gets related arrangement which is

95
Authorized licensed use limited to: Universitas Indonesia. Downloaded on November 19,2024 at 14:25:47 UTC from IEEE Xplore. Restrictions apply.
 2022 International Mobile and Embedded Technology Conference (MECON)

Architecture:

Fig. 1 State Diagram Chart

The chatbot will prompt the user for their symptoms and A. Disease Detection
as the questions progress, the symptoms will become more 1) Data Collection and Analysis
specific. Once all the symptoms of the user are entered, the We searched through many different datasets consisting of
symptoms are converted into a vector. This vector is the test diseases and symptoms. We decided to analyze 2 different
set which is then passed into the Machine Learning types of datasets, i.e., disease specific dataset and common
classification model. The model will output a disease which disease dataset and their symptoms.
the model predicts.
Heart disease Dataset

After the disease is detected, the user may opt to book an This dataset was taken from Kaggle. It consists of 15
appointment with a doctor. To do so, the bot will prompt for columns, in which 14 represent the different attributes of the
the user’s location and an API call will be made to find all the heart disease and the last columns represents weather the heart
doctors in that area which specialize in consulting for the disease is present or not.
disease which was diagnosed.

Fig. 2 Heart Disease Data Set

Analysis of Heart Disease Data Set:

96
Authorized licensed use limited to: Universitas Indonesia. Downloaded on November 19,2024 at 14:25:47 UTC from IEEE Xplore. Restrictions apply.
 2022 International Mobile and Embedded Technology Conference (MECON)

In both heart disease dataset as well as disease-symptom

dataset, the data was of the same type, i.e., categorical data.
Categorical Data requires classification machine learning
algorithms.

SVM Classifier: SVM stands for Support Vector Machine,

which is a supervised Machine Learning algorithm which
fulfills the purpose of both classification and regression
challenges. The rules generated by this algorithm will be used
further in Hybrid rule-based induction.
Fig. 3 Age Distribution

Fig. 4 Cholesterol Distribution Fig. 7 SVM Classification Analysis

Disease-Symptom Dataset

The data set we have taken was provided by Columbia Ada Boost Classifier: It is also known as Adaptive
University. It represents disease-symptom associations based Boosting. The main concept in this classifier is to allot weights
on discharge summaries of patients in New York Hospital of the classifiers and guarantee precise predictions of
Presbyterian admitted during 2004. The data set consists of uncertain observations by training the data sample in each
150 different diseases with 349 unique symptoms. iteration.

Fig. 8 AdaBoost Classification Analysis

Fig. 5 Disease-Symptom Data Set
Gradient Boosting Classifier:

Analysis of Disease-Symptom Data Set: In the modern-day world the most important aspect of a
Machine learning model is to be accurate, for this they need to
use Ensemble Techniques. One such ensemble technique is
Gradient Boosting. In gradient boosting techniques we use
many weaker models iteratively and learn from them and try
to build strong model, when we use this technique on
Classifier model it is termed as Gradient Boosting classifier.

Fig. 6 Disease and Symptom Count

2) 3.1.2. ALGORITHMS USED

97
Authorized licensed use limited to: Universitas Indonesia. Downloaded on November 19,2024 at 14:25:47 UTC from IEEE Xplore. Restrictions apply.
 2022 International Mobile and Embedded Technology Conference (MECON)

Fig. 11 Fig. 10 Decision Tree Classification Analysis

Multinomial Naïve Bayes Classifier:

This classifier based on Bayesian approach which is

famous in NLP. The algorithm gives a tag to the text and later
the likeliness of the tag is calculated for the given text and the
tag with the greatest chance is given as output.
This is very simple to use and it can take in huge amount
Fig. 9 Gradient Boosting Classification Analysis of data, therefore is very good to use in real time applications.
Random Forest Classifier: Random forests, also known as
random decision forests, are an ensemble learning approach
for classification, regression, and other problems that works
by training a large number of decision trees.

Fig. 12 Multinomial Naïve Bayes Classification Analysis

For the heart disease dataset, Support Vector Classifier

gave the best accuracy out of the 4 algorithms we used.
For the Disease-Symptom dataset, Multinomial Naïve
Bayes Classifier gave a higher accuracy than that of Decision
Tree Classifier.

Fig. 10 Random Forest Classification Analysis B. Booking A Doctor’s Appointment

A dataset consisting of 12 different doctors based on their
specializations has been taken by us. The data set consists of
the name of the doctor, specialization, timings, days and
Decision Tree Classifier: contact number.

Fig. 13 Doctor Data Set

98
Authorized licensed use limited to: Universitas Indonesia. Downloaded on November 19,2024 at 14:25:47 UTC from IEEE Xplore. Restrictions apply.
 2022 International Mobile and Embedded Technology Conference (MECON)

C. Medical Library 5. The tracker then again records the action specified
After the chatbot outputs the disease prediction of the user, by the policy.
the user will have follow-up questions. The user may ask 6. The chatbot sends a response based on the action.
about symptoms, treatments, doctors, and all types of
information specific to the disease. The user might also just IV. CONCLUSION
want to know about a disease not related to their illness. Artificial Intelligence has started to claim its role into
Creating a database which consists of all the information every aspect of our lives. Almost all industries are starting to
about a large number of diseases which are very common will use or depend on AI. Healthcare is a particular field where
help the chatbot to resolve more user queries and allow it to Artificial Advancements have not only pushed research
be more useful. forward but also saved people’s lives. Creating applications
Data of the diseases will be scraped from the medical and tools which help in the automation of important tasks,
library of ‘ada health’ and ‘Wikipedia’ and stored in a like disease detecting chatbot, can be beneficial to not only
database. patients but also doctors. Conversational AI has also entered
our lives with voice assistants like Siri, Google Home and
D. Human Like Interaction Alexa. Other than voice assistants, chatbots have started to
All chatbot developers have one common feature in mind rise in popularity thanks to the advancements in NLP. The
and that is for their chatbots to mimic a human interaction as main approach of the chat bot system is to detect and display
close as possible. Developments in the NLP and neural the disease based on the user’s symptoms with great accuracy
networks have made it possible to apply artificial intelligence and precision. The ease of using chatbots will ensure that the
in conversational tools. RASA is a tool which allows us to user does not ignore their smaller symptoms since going to
use these advancements to our advantage and create state of consult a doctor can sometimes become a great hassle. Our
the art chatbots. aim is to create an easy, convenient and accurate method to
get an idea of the user’s illness without the visiting a doctor
1) 3.4.1. RASA or going endlessly through google.
Rasa is a powerful open-source framework which works
on the principals of artificial intelligence to create industry V. REFERENCES
grade chatbots. Rasa can work with databases, APIs, [1] [1] Doctor Chatbot: Heart Disease Prediction System 1Sherwin
machine learning models, etc. fernandes, 2Rutvij Gawas, 3Preston Alvares, 4Macklon Fernandes,
5Deepmala Kale and 6Shailendra AswaleDepartment of Computer
We can partition the architecture of rasa into 2 modules Engineering, S.R.I.E.I.T, Shiroda, Goa, India
[2] [2] Chatbot System for Healthcare using Artificial Intelligence
2) [Link]. RASA NLU 1Aishwarya Kedar, 2Jyoti Dahale, 3Khushboo Patel, 4Shivani
NLU is a subset of NLP, which along with converting text Lahamage, 5Prof. S. G. Chordiya
into its semantic parts, also interprets what the user is saying. [3] [3] Medical Chatbot Techniques: A Review Andrew Reyner Wibowo
The Rasa NLU tries to understand the input to find the intents Tjiptomongsoguno, Audrey Chen, Hubert Michael Sanyoto, Edy
and extract the entities Irwansyah(B) , and Bayu Kanigoro School of Computer Science, Bina
Nusantara University, Jakarta, Indonesia
3) [Link]. RASA CORE [4] [4] A Chatbot for Psychiatric Counseling in Mental Healthcare Service
Rasa core is the dialogue management system. It decides Based on Emotional Dialogue Analysis and Sentence Generation";
what response or what action needs to be done next. 2375-0324/17 $31.00 ©2017 IEEE, 2017 IEEE 18th International
Conference on Mobile Data Management. Kyo-Joong Oh, DongKun
Lee, ByungSoo Ko, Ho-Jin Choi;
[5] [5] "Chatbot Utilization for Medical Consultant System";978-1-5386-
7573-1/18/$31.00 ©2018 IEEE, The 2018 Technology Innovation
Management and Engineering Science International Conference
(TIMES-iCON2018). Nudtaporn Rosruen and Taweesak Samanchuen;
[6] [6] Self-Diagnosing HealthCare Chatbot using Machine Learning S.
Anil Kumar, 2C. Vamsi Krishna, 3P. Nikhila Reddy, 4B. Rohith
Kumar Reddy,5I. Jeena Jacob Department of Computer Science and
Engineering, GITAM School of Technology, GITAM University-
Bengaluru Campus, Karnataka, India
[7] [7] A Pediatric Generic MedicineConsultant Chatbot “, Journalof
Automation and Control Engineering Vol. 3 BenildaEleonor V.
Comendador, Bien Michael B. Francisco, Jefferson S. Medenilla,
Fig. 14 Rasa Conversational AI Flow SharleenMae T. Nacion, and Timothy Bryle E. Serac, “Pharmabot:
, No. 2, April2015. Gillian Cameron, David Cameron, Gavin
Megaw,Raymond Bond,, SiobhanO‟Neill, Cherie Armour, Michael
The steps of the Rasa conversation AI are: McTear, “Towardsa chatbot for digitalcounselling”, Journalof
Medical Internet Research, 4(1),
1. User input is received and forwarded to the [8] [8] Chatbot for Healthcare System Using Artificial Intelligence Lekha
interpreter which converts the text into entities and Athota Student, Bachelor of Science Information Technology Amity
University, Dubai, UAE Vinod Kumar Shukla Department of
intents which are discovered by Rasa NLU. Engineering and Architecture Amity University, Dubai, UAE Nitin
2. The Tracker objects maintains the current state of Pandey Amity Institute of Information Technology, Amity University
Noida, UP, India
the conversation.
[9] [9] N-gram Accuracy Analysis in the Method of Chatbot Response,
3. The current state is passed to the policy. International Journal of Engineering & Technology. (2018)
4. The policy decides the chatbot’s next move. [10] [10] ). An intelligent webbased voice chat bot. 386 -
391.10.1109/EURCON.2009.5167660 Du Preez, S.J. & Lall, Manoj &
Sinha, S. (2009

99
Authorized licensed use limited to: Universitas Indonesia. Downloaded on November 19,2024 at 14:25:47 UTC from IEEE Xplore. Restrictions apply.
 2022 International Mobile and Embedded Technology Conference (MECON)

[11] [11] Medical Chatbot for Disease Prediction using Machine Learning
Mrs. Asha Rani K P1 , Mrs. Asha K N2 , Ranjith Kumar D R3 , Rohan4
, Ranjan Raj5 , Rohini6 Assistant Professor , Department of Computer
Science and Engineering, Dr. Ambedkar Institute of Technology,
Bangalore, India 3,4,5,6Student, Department of Computer Science and
Engineering, Dr. Ambedkar Institute of Technology, Bangalore, India
[12] [12] "A Self-Diagnosis Medical Chatbot Using Artificial Intelligence",
J. Web Dev. Web Des., vol. 3, no. 1, pp. 1-7, 2018. S. Divya, V,Divya
Madhu, Indumathi, S. Ishwarya, M. Priyasankari, S. Kalpana Devi,
[13] [13] “Mobile based healthcare management using artificial
intelligence” in International Conference on Technologies for
Sustainable Development (ICTSD), 2015, 4-6 Feb, 2015. Amiya
Kumar Tripathy , Rebeck Carvalho , Keshav Pawaskar , Suraj Yadav,
[14] [14] Chatbot for Disease Prediction and Treatment Recommendation
using Machine Learning Rohit Binu Mathew, Sandra Varghese, Sera
Elsa Joy, Swanthana Susan Alex Department of Computer Science and
Engineering Saintgits College of Engineering Kottayam, Kerala, India
[15] [15] “Natural language processing future”, 2013 International
Conference on Optical Imaging Sensor and Security (ICOSS).
doi:10.1109/icoiss.2013.6678407 M. Chandhana Surabhi.
[16] [16] "Chatbot for university related FAQs", 2017 Int. Conf. Adv.
Comput. Commun. Informatics ICACCI 2017, pp. 1525-1530,
September 2017. R. Ranoliya, N. Raghuwanshi, S. Singh,
[17] [17] “Mobile Clinical Decision Support System for Acid-base Balance
Diagnosis and Treatment Recommendation” in Acta Informatica
Medica, vol. 25, no. 2, pp. 121–125, 2017 M. Mandzuka et al.,
[18] [18] "Exploring temporal patterns of suicidal behavior on Twitter", in
Proc. IEEE Int. Conf. Healthcare Informat. Workshop (ICHI-W), Jun.
2018, pp. 55-56. J. Luo, J. Du, C. Tao, H. Xu, and Y. Zhang,
[19] [19] Natural Language Processing Enabled Cognitive Disease
Prediction Model for Varied Medical Records Implemented over ML
Techniques Vikas Kamra Amity University Uttar Pradesh Noida, India
Praveen Kumar CSE Department Masoud Mohammadian Faculty of
Science and Technology
[20] [20] Divya Madhu, Neeraj Jain C. J, Elmy Sebastain, Shinoy Shaji,
Anandhu Ajayakumar. A Novel Approach for Medical Assistance
Using Trained Chatbot, International Conference on Inventive
Communication and Computational Technologies (ICICCT

100
Authorized licensed use limited to: Universitas Indonesia. Downloaded on November 19,2024 at 14:25:47 UTC from IEEE Xplore. Restrictions apply.