UNIVERSITY INSTITUTE OF TECHNOLOGY

BARKATULLAH UNIVERSITY, BHOPAL

DEPARTMENT OF Computer Science and Engineering
YEAR 2019-2023

SYNOPSIS
of the Major Project on
Predicting - “Attention Deficit Hyperactive Disorder”

SUBMITTED IN THE PARTIAL FULFILMENT OF REQUIREMENT FOR THE

AWARD OF DEGREE OF
BACHELOR OF ENGINEERING IN COMPUTER SCIENCE AND ENGINEERING
YEAR 2022-23
UNIVERSITY INSTITUTE OF TECHNOLOGY
BARKATULLAH UNIVERSITY, BHOPAL

SUBMITTED BY
Ritu Bajpai
BE, CSE 7th semester.

UNDER THE GUIDANCE OF

Mr. Bhawani Singh Rathore Mrs. Kamini Maheshwar Dr. Divakar Singh
Project coordinator Project coordinator Head of Department

CSE, UIT-BU, Bhopal CSE, UIT-BU, Bhopal CSE, UIT-BU, Bhopal

1
 CONTENT

1. Introduction
2. Motivation
3. Objective(s) and Scope
4. Problem Identification
5. Literature Review
6. Proposed methodology/Plan of work
7. Expected outcome of the proposed work
8. References

2
1. Introduction

Attention Deficit Hyperactivity Disorder (ADHD) is a common behavioural neurodevelopment

disorder in children and adolescents and may have lifelong effects. Investigations show that
about 5% of children and 2.5% of adults suffer from this disorder. ADHD is associated with
inattention, hyperactivity, and impulsive behaviours or a combination of these symptoms and
may accompany a psychological problem such as depression or bipolar disorder. These
symptoms may increase the risk of criminal behaviour, divorce, unemployment, learning
disabilities, low social achievement, and suicide in adults. Early diagnosis of ADHD can prevent
it from adversely affecting the patients’ life. Diagnosis of ADHD is usually made based on the
information given by the person, school teachers, parents, and questionnaires. The doctor’s
experience impacts this subjective diagnosis. Regarding these issues, diagnosing ADHD disorder
is a challenging task, and the probability of misdiagnosis is high. Given that ADHD is directly
related to brain function, many studies have focused on bio-feedbacks to diagnose ADHD
objectively and help the procedure.

There are generally two categories of methods for recording bio-feedbacks: invasive and non-
invasive. Most studies utilise non-invasive bio-feedbacks like EEG, fNIRS, fMRI, EOG, and
MEG to diagnose ADHD. For example, uses the fMRI signals in the ADHD-200 dataset and
achieves an accuracy of 71.3%. In another study, investigated the relationship between the
amplitude of the hemodynamic response and ADHD patients using fNIRS data. EEG signal
recorders are inexpensive and portable, with no medical risks. In addition, the non- invasiveness
and high temporal resolution of these signals make them preferable to other bio-feedbacks in the
field of neuroscience.

EEG is a method of recording the electrical activity of the human brain using electrodes placed
on the human scalp. Hans Berger used EEG signals on humans in 1924 for the first time. These
signals span a wide range of frequency spectrums. In general, there are five frequency bands for
analysis of EEG signals: Delta (<4 Hz), Theta (4–8 Hz), Alpha (8–13 Hz), Beta (13–30 Hz), and
Gamma (greater than30 Hz). Each of these bands has its unique biological characteristic. They
have various applications, such as diagnosing sleep problems, cancer, tumours, and other brain-
related disorders. JF Lubar used EEG bio-feedback for the first time to diagnose ADHD. In his

3
study, he observed that event- related potentials (ERPs) in ADHD patients are different from the
control group. In his 15-year study, he found that patients with ADHD had increased theta
activity and decreased beta power.

Fig. 1 Characteristic of ML

To analyse the EEG data, we first render the preprocessing step. The purpose of preprocessing is
to remove internal and external artefacts from the EEG data. In most studies, preprocessing is
carried out manually, which requires expert knowledge. Automated methods such as
Independent Component Analysis (ICA), and machine learning based approach, are among other
techniques to remove these artefacts. The next step is to prepare the input for the classifier
model. There are different input formulations for this task. Common Spatial Patterns (SCP)
Discrete Wavelet Transform (DWT) Lyapunov Exponent (LE) and Fractal Dimension (FD) are
among the most common feature extraction methods. Some studies use features such as
Spectrograms Fast Fourier Transform maps and 3D Grids.

It is hard to detect high-level patterns in the complex data extracted from the human brain. For
years, machine learning has been developing methods for recognising patterns with the objective
to predict and classify brain data. Machine learning algorithms like Support Vector Machine
(SVM), K-Nearest Neighbor (KNN), and Feed forward neural network are applied to diagnose
ADHD patients utilising EEG data. As a consequence of the increasing computing power of
today’s processors and the high performance of neural network models, these models have been

4
used frequently in recent years in various fields, and basic methods such as SVM and KNN are

less favourable.

Neural networks are a type of machine learning algorithm that at- tempts to mimic the process of
the human brain. Neural networks were first introduced in 1943. The computational load of the
neural network distributes to cells called neuron. These cells are in a network structure
consisting of three layers: input, hidden, and output, and they send the results of their
computations on the data through their connections with other cells to carry out the learning
process and obtain the coefficients needed to predict the class of new samples. Many studies
have used Multi-Layer Perceptron (MLP) to classify ADHD patients and control groups.

Deep neural networks (DNNs) are a type of neural network known to have multiple layers in
their network structure. In recent years, deep learning has been an area of interest for clinical
diagnosis. Due to the end-to-end architecture and the power of deep neural networks to extract
representation from the data, they do not need a separate feature extraction phase. Convolutional
neural networks (CNNs) are deep neural networks with convolutional layers. Due to the success
of this model in the field of machine vision, researchers are interested in this type of neural
network in clinical applications as well. For example, first processes the EEG data and then uses
the wavelet transform to ac- quire the spectrogram. Finally, these spectrograms are input to a
CNN with four layers of filtering and pooling. The model accuracy in this paper is 88%, and it is
superior to similar tasks since it does not require manual channel selection. extracts a 3-channel
RGB colour image from different bands of EEG signal and uses this transformed input to a
three-layer CNN with Max-pooling and dropout. Their model achieved an accuracy of 98.48.

5
2. Motivation

To enhance service performance in different sector, IOT network increases the flexibility. This
invites attacker to take advantage of network vulnerability. Many of existing systems have
security flaws that render them susceptible to intrusions, penetrations, and other forms of attacks.
So protection from such type of IOT network attacks need a model that help in detection and
alarming. Most of them depends on the input feature set and predict the type of session either
normal or attacked. This limitation need to be enhanced as type of attack help to identify damage
files. The future attacks on any system, network is controlled and stopped by analysing attack
feature samples and behaviour. The wide variety and quantity of now-a-day systems and the
amount of sensitive information they store provide numerous opportunities to illegally make a
profit out of subverting legitimate systems. Analysis time of the system was also high as this
increase network and processing load. These issues need to be resolved as people are more
dependent on digital platforms, network, computers, etc.

6
3. Objective(s) and Scope

The main objective of this project can be broken down into following points:
i. To study the application of machine learning in health care systems.
ii. To identify the Attention Deficit Hyperactivity Disorder (ADHD) in early stages.
iii. To identify the Attention Deficit Hyperactivity Disorder (ADHD) in most frequent age
groups.
iv. To develop a machine learning model for identifying ADHD.
v. Compare the developed model on various evaluation parameters on different testing

dataset size.

7
4. Problem Identification

It was obtained from the literature survey that most of scholars have proposed Attention Deficit
Hyperactivity Disorder (ADHD) using machine learning or deep learning separately but not used
both together as assembly learning. It was also found that most of the scholars work in individual
class detection first was machine learning and the other was deep learning. Hence it is highly
required to develop a detection model that identify Attention Deficit Hyperactivity Disorder
(ADHD). Input feature vector in a learning model should be optimised by feature reduction
techniques. Some of the researchers have performed this work but a scholar work in two class
detection first was normal and the other was malicious. Detection model accuracy of Attention
Deficit Hyperactivity Disorder (ADHD) was high which should be reduced with a reduction in
detection time of work.

8
5. Literature Review

Mohammad Reza Mohammadi had propoded the research paper in The Korean Society of
Medical & Biological Engineering and Springer 2016 with the accuracy of 92.28% and 93.65%
were achieved using mRMR method and DISR method using MLP, respectively.

Mehdi Samavati had proposed this in 20th Iranian Conference on Electrical Engineering,
(ICEE2012) the accuracy of 92.28% usingConvolutional Neural Network (APSO-CNN).

9
6. Proposed methodology
This work focus on developing a model that identify Attention Deficit Hyperactivity Disorder
(ADHD). For this input dataset is pass from the pre-processing step that cleans the data for
finding the feature set. Feature identification is done by the dynamic approach with soft
computing algorithms. Selected feature set values were further process for the training of
machine learning algorithms & deep learning algorithms. Trained neural network is tested
against various data. Working steps were shown in fig.3.

Training Dataset

Pre-Processing

Feature Detection

Training Feature Set Desired Output Class

Deep Learning

Fig. 3 Block diagram of proposed IOT network attack detection model..

Dataset: ADHD dataset was taken from IEEE.

10
Pre-Processing: In this step of proposed model input dataset is transform into matrix format.
Incomplete session is also removed in this step. For increasing the training efficiency of learning
model type of session class in training dataset will be balance.

Feature Detection:
Genetic algorithm was used in the work for the selection of features for identifying the class
representative sessions. These algorithms are dynamic not need any training and guidance. As
input datasets have any number of features that need adoptable algorithm. So this work will find
feature set from genetic soft computing approach.

Machine Learning and Deep Learning:

To train a machine learning & deep neural network filtered session are pass as input vector and
corresponding class act as desired output. Features from the sessions were filter to reduce the
confusion while training the neural network. Sigmoid function will be used for the activation of
neuron. As per desired output class value neuron weight values will be adjust for learning of
neural network. Work will find the feature pattern as per subclass of attack.

Experimental Work: Implementation and comparing of proposed model was done on Google

Colab software. Experimental machine having 4 GB ram, i3 6th generation processor.

11
7. Expected outcome of the proposed work

Attention Deficit Hyperactivity Disorder (ADHD) is an important issue as number of

researchers have proposed different model for its detection. The early detection of Attention
Deficit Hyperactivity Disorder (ADHD), can be cured using medical sciences. This work will
develop a Attention Deficit Hyperactivity Disorder (ADHD) model. Find good set of features
from dataset that impacts the detection accuracy of true alarm generation. The proposed model
will detect Attention Deficit Hyperactivity Disorder (ADHD) with machine learning & deep
learning model with an objective to identify its subclass as well.

12
8. References

1. Louis Columbus. Roundup Of Internet Of Things Forecasts And Market Estimates, 2016.
2. Wei Zhou et al. “The effect of iot new features on security and privacy: New threats,
existing solutions, and challenges yet to be solved”. In: IEEE Internet of Things Journal
6.2 (2018).
3. The Independent. Hackers Now Able to Take Control of Cars to Cause Deliberate
Accidents, Scientists Warn. 2017 (accessed June 24, 2020).
4. International Data Corporation (IDC). The Growth in Connected IoT Devices Is
Expected to Generate 79.4ZB of Data in 2025, According to a New IDC Forecast. 2019.
5. Mario Frustaci et al. “Evaluating critical security issues of the IoT world: Present and
future challenges”. In: IEEE Internet of things journal 5.4 (2017).
6. Kamarularifin Abd Jalill, MohamadNoormanMasrek “Comparison of Machine Learning
Algorithms Performance in Detecting Network Intrusion” 201O International Conference

on Networking and Information Technology 978-1-4244-7578-0/$26.00 © 2010 IEEE

13