Classification of Alzheimer's Disease using Machine Learning

Techniques

Muhammad Shahbaz

, Shahzad Ali

2 a

, Aziz Guergachi

, Aneeta Niazi

and Amina Umer

Department of Computer Science and Engineering, University of Engineering and Technology, Lahore-54890, Pakistan

Department of Information Technology, University of Education Lahore, Multan Campus, Multan, Pakistan

Department of Information Technology Management, TRS, Ryerson University, Toronto, ON, Canada

aminaumer@gmail.com

Keywords: Alzheimer’s Disease, Data Mining, Machine Learning, Healthcare, Classification, Neurodegeneration.

Abstract: Alzheimer's disease (AD) is a commonly known and widespread neurodegenerative disease which causes

cognitive impairment. Although in medicine and healthcare areas, it is one of the frequently studied diseases

of the nervous system despite that it has no cure or any way to slow or stop its progression. However, there

are different options (drug or non-drug options) that may help to treat symptoms of the AD at its different

stages to improve the patient’s quality of life. As the AD progresses with time, the patients at its different

stages need to be treated differently. For that purpose, the early detection and classification of the stages of

the AD can be very helpful for the treatment of symptoms of the disease. On the other hand, the use of

computing resources in healthcare departments is continuously increasing and it is becoming the norm to

record the patient’ data electronically that was traditionally recorded on paper-based forms. This yield

increased access to a large number of electronic health records (EHRs). Machine learning, and data mining

techniques can be applied to these EHRs to enhance the quality and productivity of medicine and healthcare

centers. In this paper, six different machine learning and data mining algorithms including k-nearest

neighbours (k-NN), decision tree (DT), rule induction, Naive Bayes, generalized linear model (GLM) and

deep learning algorithm are applied on the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset in

order to classify the five different stages of the AD and to identify the most distinguishing attribute for each

stage of the AD among ADNI dataset. The results of the study revealed that the GLM can efficiently classify

the stages of the AD with an accuracy of 88.24% on the test dataset. The results also revealed these techniques

can be successfully used in medicine and healthcare for the early detection and diagnosis of the disease.

1 INTRODUCTION

Neurodegenerative, continuous deterioration of

neurons, diseases are usually considered as a group of

disorders that damage the working competence of the

human nervous system intensely and progressively

(Scatena et al., 2007). Alzheimer’s disease (AD) is

the key public health concern throughout the world

and one of the most widespread neurodegenerative

disorder (Small, 2005). The AD is a cureless disease

because it has no diagnosis and treatment methods to

slow its progression or stop its onset (Unay et al.,

2010). The median survival duration of the patients

suffering from AD has been estimated to be only 3.1

years for the initial diagnosis of the probable AD, and

https://orcid.org/0000-0002-0608-9515

only 3.5 years for the initial diagnosis of possible AD

(Wolfson et al., 2001).

The frequency of AD occurrence is becoming

more common in late life, especially the people at the

age of greater than 65 are at high risk (Cummings et

al., 2014). At present, there are approximately 44

million victims of AD dementia throughout the

world. If the breakthroughs fail to identify the

prevention and diagnosis of the AD, it is expected to

be increased to a number of more than 100 million by

the year 2050 (Association et al., 2013; Touhy et al.,

2014). During the AD, degeneration of the brain

progresses with time. Therefore, the patients

suffering from the AD should be categorized into

different subgroups, depending upon the stage of the

296

Shahbaz, M., Ali, S., Guergachi, A., Niazi, A. and Umer, A.

Classiﬁcation of Alzheimer’s Disease using Machine Learning Techniques.

DOI: 10.5220/0007949902960303

In Proceedings of the 8th International Conference on Data Science, Technology and Applications (DATA 2019), pages 296-303

ISBN: 978-989-758-377-3

disease. This division is critical because the patients

at different stages of the AD are required to be treated

differently, and the same medication cannot be used

for all of them (Gamberger et al., 2017). For that

purpose, the classification of different stages of the

AD can be very helpful for the treatment of symptoms

of the disease to improve the patient’s quality of life.

The use of computing resources in healthcare

departments is continuously increasing and it is

becoming the norm to record the patient data

electronically that was traditionally recorded on

paper-based forms. This yield increased access to

many electronic health records (EHRs) but 80% of

the data is unstructured. As a result, the processing of

the unstructured data is challenging and difficult

using database management software and other

traditional methods. Machine learning and data

mining tools and techniques can be applied to these

EHRs to enhance the quality and productivity of

medicine and healthcare centers (Alonso et al., 2018).

Data mining or knowledge discovery is the practice

of finding the unknown and useful patterns from

many pre-existing datasets. Such patterns are used to

understand the historical dataset, to classify new data

and generate summaries of data. In this way, data

mining helps in the discovery of the deeper patterns

in data, as well as classify or group records on the

basis of similarities or dissimilarities between them

(Sumathi and Sivanandam, 2006).

Since the last few decades, data mining has been

extensively used in many areas such as marketing,

retail, banking, stock market prediction, and medicine

and healthcare, etc. (Agarwal et al., 2018; Canlas,

2009; Thenmozhi and Deepika, 2014; Yang et al.,

2018). The algorithms used for data mining can be

applied to group different subjects based on the

similarities in their attributes (Eapen, 2004). Many

studies (Soni and Gandhi, 2010; Dipnall et al., 2016;

Ni et al., 2014 have shown that machine learning, and

data mining techniques have been significantly used

in medicine and healthcare research. The application

of data mining in medicine helps in the efficient

prognosis of several diseases, understanding the

classification of disease, specifically in neuroscience,

and biomedicine. The main aim of this research work

is to apply machine learning and data mining

techniques on the Alzheimer's Disease Neuroimaging

Initiative (ADNI) dataset to classify the different

stages of the AD. Moreover, the sub-objective of this

research work is to identify the most distinguishing

attribute for each of the different stages of the AD

among ADNI dataset.

2 DATA

2.1 Data Description

In this research work, the dataset has been collected

from The Alzheimer’s Disease Prediction of

Longitudinal Evolution (TADPOLE) challenge

which is available at https://tadpole.grand-

challenge.org. It is the dataset recorded from the

North American individuals, who participated in

ADNI. ADNI is the multicentre study aimed to

improve the clinical, genetic, biochemical, and

imaging biomarkers for early diagnosis of the AD. A

standard set of procedures and protocols is followed

during ADNI data collection, to avoid any

inconsistencies in the data (Weiner et al., 2013). The

TADPOLE data has been recorded for both male and

female participants, including mild cognitive

impairment subjects, old aged individuals, and AD

patients. The data contain records of participants’

examination carried out at 62 different sites, at

different monthly intervals, ranging from baseline

(i.e. 0 months) to 120 months, from July 2005 to May

2017.

2.2 Data Exploration

The original TADPOLE dataset contains 1,907

attributes for 1,737 participants. On the basis of

diagnosis, these participants have been divided into

five different classes namely Cognitively Normal

(CN), Early Mild Cognitive Impairment (EMCI),

Late Mild Cognitive Impairment (LMCI), Subjective

Memory Complaint (SMC), and Alzheimer’s Disease

(AD). The number of participants and their

examination records for each of the five stages of AD

included in the TADPOLE dataset are illustrated in

Table 1.

Table 1: Original TADPOLE dataset details.

Class

No. of Participants

No. of Records

417

3,821

EMCI

310

2,319

LMCI

562

4,644

SMC

106

389

342

1,568

TADPOLE dataset contains 12,741 examination

records of 1,737 participants. The dataset is unevenly

distributed among the five classes of participants, i.e.

LMCI and CN have more data as compared to AD,

EMCI, and SMC. The class SMC has the least data,

with just 106 participants having 389 examination

records (Table 1).

Classiﬁcation of Alzheimer’s Disease using Machine Learning Techniques

297

2.3 Data Pre-Processing and Feature

Selection

In this research work, a subset of TADPOLE dataset

has been used. The examination records of 530

participants (106 participants from each AD stage)

have been selected out of 1,737 participants, to

perform experiments using different data mining

algorithms for the classification of five different

stages of the AD. In data mining techniques, the

completeness of data is very critical for precise

modeling. While the TADPOLE dataset contains a lot

of sparseness. Therefore, we have selected only those

attributes, which have maximum data coverage. An

attribute will have complete data coverage if it will

have a value for all the 12,741 instances of

examination records. But, there are only 41 attributes,

out of 1,907 attributes, which have a value for more

than 10,000 instances of examination records.

Furthermore, an analysis has been carried out

to remove redundancy from the 41 attributes having

enough data. There are 6 attributes in the TADPOLE

dataset, that indicate the time of participant’s

examination, including EXAMDATE (date of

examination), EXAMDATE_bl (date of first

examination), M (months of examination, to nearest

6 months as continuous), Month (months of

examination, to nearest 6 months as a factor),

Month_bl (fractional value of months of

examination), and VISCODE (participant’s visit

code, indicating months of examination). As all these

attributes indicate the time of the participant’s

examination. In this way, only one attribute namely

VISCODE has been selected for indicating the

participant’s examination date, and all other five

redundant attributes have been removed. Similarly,

the dataset contains two attributes for the participant’s

identification, including RID (participant Roaster ID)

and PTID (Participant ID). The attribute PTID has

been removed from the dataset to avoid redundant

IDs. After data analysis and pre-processing, 28

attributes have been selected for this investigation.

These attributes are divided into three

categories. The first category is demographics

attributes which include the general quantifiable

characteristics of participants. The second category is

cognitive assessment attributes which include the

attributes representing the cognitive behavior of a

participant. For that purpose, different cognitive

assessment tests are carried out and scores are

assigned to the participant based on their cognitive

abilities. Finally, the third category is clinical

assessment attributes which include the significant

biomarkers of the AD. These attributes have been

recorded after the clinical examination of all

participants. The list of demographic, cognitive

assessment and clinical assessment attributes with

their description that have been extracted for analysis

is illustrated in Table 2, 3 and 4, respectively.

Table 2: List of demographic attributes in the dataset used in this analysis.

Label

Description

Data Type

Units

RID

Participant’s Roaster ID.

Numeric

AGE

Participant’s age.

Numeric

Years

PTGENDER

Participant’s gender

Nominal

PTEDUCAT

Participant’s education

Numeric

Years

PTETHCAT

Participant’s ethnicity

Nominal

PTRACCAT

Participant’s race

Nominal

PTMARRY

Participant’s marital status.

Nominal

VISCODE

Participant’s Visit code.

Nominal

Years_bl

Participant’s year of examination.

Numeric

Years

SITE

A code indicating the site of a participant’s examination

Numeric

Table 3: List of cognitive assessment attributes in the dataset used in this analysis.

Label

Description

Data Type

Units

CDRSB_bl

Clinical Dementia Rating Sum of Boxes (core)

Numeric

ADAS11_bl

11 item-AD Cognitive Scale (score)

Numeric

ADAS13_bl

13 item-AD Cognitive Scale (score)

Numeric

MMSE_bl

Mini-Mental State Examination (score)

Numeric

RAVLT_immediate_bl,

RAVLT_learning_bl,

RAVLT_forgetting_bl,

RAVLT_perc_forgetting_bl

Rey's Auditory Verbal Learning Test (scores for immediate

response, learning, forgetting and percentage forgetting)

Numeric

FAQ_bl

Functional Activities Questionnaire

Numeric

DATA 2019 - 8th International Conference on Data Science, Technology and Applications

298

Table 4: List of clinical assessment attributes in the dataset used in this analysis.

Label

Description

Data Type

Units

APOE4

APOE4 gene presence

Binary

Hippocampus_bl

Volume of hippocampus

Numeric

Ventricles_bl

Volume of ventricles

Numeric

WholeBrain_bl

volume of Brain

Numeric

Fusiform_bl

The volume of the fusiform gyrus.

Numeric

Entorhinal_bl

The volume of the entorhinal cortex.

Numeric

MidTemp_bl

The volume of the middle temporal gyrus.

Numeric

ICV

Intra Cranial Volume

Numeric

2.4 Data Partitioning

The dataset is divided into a training dataset (70% of

the entire dataset) and test datasets (30% of the entire

dataset). According to this division, the 2,164

examination records are used for training of the data

mining models, and the 927 examination records are

used for model testing. The data counts of training

and test dataset after pre-processing are given in

Table 5. It illustrates that all the five classes have a

different number of examination records. Because in

TADPOLE dataset, the number of examination

records is not the same for all participants.

Table 5: Data Counts of each AD stage for the Training and

Test Dataset.

AD Stage

Training Samples

Test Samples

632

270

EMCI

448

192

LMCI

566

243

SMC

217

301

129

3 METHODOLOGY

In this research work, six different machine learning

and data mining algorithms including K-nearest

neighbors (K-NN), decision tree (DT), rule induction,

Naive Bayes, generalized linear model (GLM) and

deep learning algorithm are applied on ADNI dataset

in order to classify the five different stages of the AD.

In this investigation, rapidminer studio, one of the

famous data mining tools, is used for implementing

all these algorithms.

3.1 K-Nearest Neighbours (K-NN)

K-nearest neighbor algorithm is a simple data mining

technique used for both classification and regression

problems. K-NN classification algorithm assigns an

object to a specific class based on the majority classes

of its K neighbors. The value of K, a positive integer,

defines the number of neighbors to be considered for

polling (Zhang and Zhou, 2005). In this analysis, the

value of K is set to 11 which is selected using the trial

and error method.

3.2 Decision Tree (DT)

The decision tree algorithm is a predictive modeling

technique commonly used for classification in data

mining, statistics, and machine learning applications.

It classifies the dataset by computing the information

gain values for all attributes of a dataset. The leaf

nodes in a tree denote a class label while the branches

to these leaf nodes denote the combination of input

variables that lead to those class labels (Shahbaz et

al., 2013).

3.3 Rule Induction

The rule induction is a data mining algorithm, in

which a pruned set of rules is extracted from the

training data, based on maximum values of

information gain. The rules are in the form of ‘if-then’

statements (see Appendix B). Rule sets have an

advantage over decision trees, as they are simpler to

comprehend and can be represented in first-order

logic (Stepanova et al., 2018).

3.4 Naive Bayes Algorithm

The Naive Bayes algorithm is one of the data mining

and machine learning classification technique which

is based on the Bayesian theorem. It used to find the

probability of an attribute based on other known

probabilities that are related to the attribute. It uses

Gaussian probability densities for data modeling

(Thomas and Princy, 2016).

Classiﬁcation of Alzheimer’s Disease using Machine Learning Techniques

299

3.5 Generalized Linear Model (GLM)

The generalized linear model (GLM) is a supervised

machine learning approach used for both

classification and regression problems. It is an

extension of traditional linear models. GLM classifies

the data based on the maximum likelihood between

attributes. It performs parallel computations and is an

extremely fast machine learning approach and works

very well for models with a limited number of

predictors (Guisan et al., 2002).

3.6 Deep Learning

Deep learning models are one of the machine learning

techniques. It is based on the multilayer feedforward

artificial neural networks, which are vaguely inspired

by the working of the biological nervous system or

the human brain. In this investigation, a deep neural

network with two hidden layers is trained using

backpropagation learning algorithm. The detailed

working of deep neural networks is given in (Levine

et al., 2018). The schematic representation of the

proposed methodology is illustrated in Figure 1.

Figure 1: Schematic representation of the proposed

methodology.

3.7 Model Evaluation – Confusion

Matrix

The performance of the classifier is usually evaluated

using a confusion matrix. It is a specific table which

is used to presents the true classes and classifier

predicted classes and illustrates the type of errors

made by the classifier. Confusion matrix for binary

classification (class ‘0’ and class ‘1’) is shown in Fig.

2. There are four different terminologies used in the

confusion matrix which are given as follows:

 True Positive (TP) means classifier predicted

positive (‘1’) and it is true (‘1’).

 True Negative (TN) means classifier

predicted negative (‘0’) and it is true (‘0’).

 False Positive (FP) means classifier predicted

positive (‘1’) and it is false (‘0’).

 False Negative (FN) means classifier

predicted negative (‘0’) and it is false (‘1’).

Figure 2: Confusion matrix for binary classification.

Accuracy, precision, and recall are the important

performance metrics that are calculated from a

confusion matrix for a classification model.

Accuracy means how often the classification

model correctly classifies the data samples?

 





(1)

Precision is the number of TP classes over the

sum of both the TP classes and FP classes.

 





(2)

Recall is the number of TP classes over the sum

of both the TP classes and FN classes.

 





(3)

4 RESULTS AND DISCUSSIONS

In this investigation, all the classification models are

trained with the 10 folds cross-validation of models

on the training dataset. Cross-validation is performed

with the training dataset, to avoid overfitting of the

models. The performance of the classifiers is

evaluated using the unseen test dataset. The

classification accuracy of the classifiers during the

validation period and the test period is illustrated in

Table 6.

Table 6: The classification accuracy of data mining

classifiers during Validation and test period.

Classifier

Accuracy (%)

Validation

Test

K-NN

73.10

43.26

Decision Tree

76.43

74.22

Rule Induction

92.47

69.69

Naive Bayes

79.44

74.65

Generalized Linear Model

92.75

88.24

Deep Learning

78.79

78.32

DATA 2019 - 8th International Conference on Data Science, Technology and Applications

300

Table 7: Confusion matrix obtained by applying a generalized linear model to the test dataset.

True CN

True LMCI

True EMCI

True SMC

True AD

Precision

Pred. CN

255

89.16%

Pred. LMCI

209

98.12%

Pred. EMCI

174

79.45%

Pred. SMC

63.75%

Pred. AD

129

100 %

Recall

94.44%

86.01%

90.62%

54.84%

100 %

The results analysis indicates that the generalized

linear model outperforms other classifiers and gives

an accuracy of 88.24% during the test period.

Reasonable accuracies are obtained for deep learning

and Naive Bayes algorithms i.e. 78.32% and 74.65%

respectively. It can also be observed that the results

obtained on test data are quite close to the results

obtained from cross-validation period. This shows

that the developed models are not overfitted during

their training period. From the decision tree and rule

induction models, it has been observed that the most

distinguishing attribute for the five stages of the AD

is the CDRSB cognitive test, as it appears at the top

of the decision tree. The most distinguishing clinical

assessment attribute is the volume of the whole brain,

whereas the most distinguishing demographic

attribute is the age of the patient (see Appendices A

and B). The detailed results obtained by applying the

generalized linear algorithm during the test period is

illustrated in Table 7.

It can be observed that generalized linear model

correctly classified most of the unseen instances of

AD, CN, EMCI and LMCI classes, with the class

recall of 100.00%, 94.44%, 90.62%, and 86.01%

respectively, and class precision of 100.00%, 89.16%,

98.12%, and 79.45% respectively. The results of the

SMC class show misclassification of testing

instances. It can be observed from the above table that

25 out of 93 instances of SMC class have been

misclassified with the CN class. This is because the

patients belonging to SMC class have very similar

values of clinical assessment attributes as those who

belong to CN class. As well as, 28 out of 243

instances of the LMCI class have been misclassified

as EMCI. This is because the attribute values of

EMCI and LMCI classes overlap with each other.

5 CONCLUSION AND FUTURE

DIRECTIONS

Machine learning, and data mining techniques are

very helpful in medicine and healthcare studies for

early detection and diagnosis of several diseases. The

results of the research work illustrate that accuracy of

the GLM is 88.24% for the test period. The results

also indicate that the most distinguishing attributes

for the different stages of AD include the CDRSB

cognitive test among the cognitive assessment

attributes, the volume of the whole brain among the

clinical assessment attributes and age of the patient

among the demographic attributes. Furthermore, the

results proved that the machine learning, and data

mining techniques can be successfully used in the

early detection, prediction, and diagnosis of several

diseases. The accuracy of the AD stages classification

could be further improved by increasing the number

of instances for EMCI and SMC classes so that the

model can be trained with sufficient and balanced

data for all classes.

ACKNOWLEDGMENTS

The authors are thankful to the Alzheimer's Disease

Neuroimaging Initiative (ADNI) data committee for

providing access to their AD dataset.

REFERENCES

Agarwal, N., Koti, S. R., Saran, S., and Kumar, A. S.

(2018). Data mining techniques for predicting dengue

outbreak in geospatial domain using weather

parameters for new delhi, india. CURRENT SCIENCE,

114(11):2281–2291.

Alonso, S. G., De La Torre-D´ıez, I., Hamrioui, S., L´opez-

Coronado, M., Barreno, D. C., Nozaleda, L. M., and

Franco, M. (2018). Data mining algorithms and

techniques in mental health: a systematic review.

Journal of medical systems, 42(9):161.

Association, A. et al. (2013). 2013 alzheimer’s disease facts

and figures. Alzheimer’s & dementia, 9(2):208–245.

Canlas, R. (2009). Data mining in healthcare: Current

applications and issues. School of Information Systems

& Management, Carnegie Mellon University,

Australia.

Cummings, J. L., Morstorf, T., and Zhong, K. (2014).

Alzheimer’s disease drug-development pipeline: few

Classiﬁcation of Alzheimer’s Disease using Machine Learning Techniques

301

candidates, frequent failures. Alzheimer’s research &

therapy, 6(4):37.

Dipnall, J. F., Pasco, J. A., Berk, M., Williams, L. J., Dodd,

S., Jacka, F. N., and Meyer, D. (2016). Fusing data

mining, machine learning and traditional statistics to

detect biomarkers associated with depression. PloS one,

11(2): e0148195.

Eapen, A. G. (2004). Application of data mining in medical

applications. Master’s thesis, University of Waterloo.

Gamberger, D., Lavraˇc, N., Srivatsa, S., Tanzi, R. E., and

Doraiswamy, P. M. (2017). Identification of clusters of

rapid and slow decliners among subjects at risk for

alzheimer’s disease. Scientific reports, 7(1):6763.

Guisan, A., Edwards Jr, T. C., and Hastie, T. (2002).

Generalized linear and generalized additive models in

studies of species distributions: setting the scene.

Ecological modelling, 157(2-3):89–100.

Levine, S., Pastor, P., Krizhevsky, A., Ibarz, J., and Quillen,

D. (2018). Learning hand-eye coordination for robotic

grasping with deep learning and large-scale data

collection. The International Journal of Robotics

Research, 37(4-5):421–436.

Ni, H., Yang, X., Fang, C., Guo, Y., Xu, M., and He, Y.

(2014). Data mining-based study on sub-mentally

healthy state among residents in eight provinces and

cities in china. Journal of Traditional Chinese

Medicine, 34(4):511–517.

Scatena, R., Martorana, G. E., Bottoni, P., Botta, G.,

Pastore, P., and Giardina, B. (2007). An update on

pharmacological approaches to neurodegenerative

diseases. Expert opinion on investigational drugs,

16(1):59–72.

Shahbaz, M., Guergachi, A., Noreen, A., and Shaheen, M.

(2013). A data mining approach to recognize objects in

satellite images to predict natural resources. In IAENG

Transactions on Engineering Technologies, pages 215–

230. Springer.

Small, D. H. (2005). Acetylcholinesterase inhibitors for the

treatment of dementia in alzheimer’s disease: do we

need new inhibitors? Expert opinion on emerging

drugs, 10(4):817–825.

Soni, N. and Gandhi, C. (2010). Application of data mining

to health care. International Journal of Computer

Science and its Applications.

Stepanova, D., Gad-Elrab, M. H., and Ho, V. T. (2018).

Rule induction and reasoning over knowledge graphs.

In Reasoning Web International Summer School, pages

142–172. Springer.

Sumathi, S. and Sivanandam, S. (2006). Introduction to

data mining principles. Introduction to data mining and

its applications, pages 1–20.

Thenmozhi, K. and Deepika, P. (2014). Heart disease

prediction using classification with different decision

tree techniques. International Journal of Engineering

Research and General Science, 2(6):6–11.

Thomas, J. and Princy, R. T. (2016). Human heart disease

prediction system using data mining techniques. In

2016 International Conference on Circuit, Power and

Computing Technologies (ICCPCT), pages 1–5. IEEE.

Touhy, T. A., Jett, K. F., Boscart, V., and McCleary, L.

(2014). Ebersole and Hess’ Gerontological Nursing and

Healthy Aging, Canadian Edition-E-Book. Elsevier

Health Sciences.

Unay, D., Ekin, A., and Jasinschi, R. S. (2010). Local

structure-based region-of-interest retrieval in brain mr

images. IEEE Transactions on Information technology

in Biomedicine, 14(4):897–903.

Weiner, M. W., Veitch, D. P., Aisen, P. S., Beckett, L. A.,

Cairns, N. J., Green, R. C., Harvey, D., Jack, C. R.,

Jagust, W., Liu, E., et al. (2013). The alzheimer’s

disease neuroimaging initiative: a review of papers

published since its inception. Alzheimer’s & Dementia,

9(5):e111–e194.

Wolfson, C., Wolfson, D. B., Asgharian, M., M’lan, C. E.,

Østbye, T., Rockwood, K., and Hogan, D. f. (2001). A

reevaluation of the duration of survival after the onset

of dementia. New England Journal of Medicine,

344(15):1111–1116.

Yang, J., Li, J., and Xu, Q. (2018). A highly efficient big

data mining algorithm based on stock market.

International Journal of Grid and High Performance

Computing (IJGHPC), 10(2):14–33.

Zhang, M.-L. and Zhou, Z.-H. (2005). A k-nearest

neighbour based algorithm for multi-label

classification. In Granular Computing, 2005 IEEE

International Conference on, volume 2, pages 718–721.

IEEE.

APPENDIX

Decision Tree Model

CDRSB_bl>0.250

| CDRSB_bl>2.750

| | MMSE_bl>26.500

| | | AGE>75.300: LMCI {AD=0, CN=0, EMCI=0,

LMCI=13, SMC=0}

| | | AGE≤75.300: EMCI {AD=0, CN=0, EMCI=8,

LMCI=0, SMC=0}

| | MMSE_bl≤26.500: AD {AD=254, CN=0,

EMCI=5, LMCI=0, SMC=0}

| CDRSB_bl≤2.750

| | MMSE_bl>23.500

| | | RAVLT_perc_forgetting_bl>-9.167

| | | | AGE>88.850: AD {AD=5, CN=0, EMCI=0,

LMCI=0, SMC=0}

| | | | AGE≤88.850

{AD=0, CN=0, EMCI=2, LMCI=0, SMC=0}

{AD=0, CN=0, EMCI=0, LMCI=0, SMC=5}

DATA 2019 - 8th International Conference on Data Science, Technology and Applications

302

EMCI {AD=0, CN=0, EMCI=78, LMCI=0, SMC=5}

LMCI {AD=23, CN=32, EMCI=344, LMCI=553,

SMC=3}

| | | | | | | | SITE≤64.500: SMC {AD=0, CN=0,

EMCI=0, LMCI=0, SMC=4}

| | | | | | | SITE>67.500: EMCI {AD=0, CN=0,

EMCI=6, LMCI=0, SMC=0}

| | | | | | | SITE≤67.500: CN {AD=0, CN=14,

EMCI=0, LMCI=0, SMC=0}

| | | RAVLT_perc_forgetting_bl≤-9.167: SMC

{AD=0, CN=0, EMCI=0, LMCI=0, SMC=5}

| | MMSE_bl≤23.500: AD {AD=19, CN=0,

EMCI=0, LMCI=0, SMC=0}

CDRSB_bl≤0.250

| AGE>69.550

| | SITE>147: SMC {AD=0, CN=0, EMCI=0,

LMCI=0, SMC=6}

| | SITE≤147

| | | WholeBrain_bl>1234525: SMC {AD=0,

CN=0, EMCI=0, LMCI=0, SMC=6}

| | | WholeBrain_bl≤1234525

| | | | FAQ_bl>2.500: SMC {AD=0, CN=0,

EMCI=0, LMCI=0, SMC=5}

| | | | FAQ_bl≤2.500

| | | | | ADAS13_bl>19.500: SMC {AD=0, CN=0,

CN=1, EMCI=0, LMCI=0, SMC=5}

| | | | | | | | ICV_bl≤1814440: CN {AD=0,

CN=566, EMCI=0, LMCI=0, SMC=77}

{AD=0, CN=0, EMCI=0, LMCI=0, SMC=3}

| | | | | | MMSE_bl≤25: SMC {AD=0, CN=0,

EMCI=0, LMCI=0, SMC=4}

| AGE≤69.550

| | PTEDUCAT>10.500

| | | AGE>64.050: SMC {AD=0, CN=0, EMCI=0,

LMCI=0, SMC=83}

| | | AGE≤64.050

| | | | PTGENDER = Female: SMC {AD=0, CN=0,

EMCI=0, LMCI=0, SMC=2}

| | | | PTGENDER = Male: CN {AD=0, CN=12,

EMCI=0, LMCI=0, SMC=0}

| | PTEDUCAT≤10.500: CN {AD=0, CN=7,

EMCI=0, LMCI=0, SMC=0}

Rule Induction Model

if CDRSB_bl≤0.750 and Fusiform_bl≤16443.500

and ADAS13_bl≤15.500 and

RAVLT_immediate_bl>32 then CN (0/282/7/0/16)

if RAVLT_perc_forgetting_bl>86.607 and

MMSE_bl>26.500 and ADAS13_bl>19.500 then

LMCI (0/0/4/131/0)

if MidTemp_bl>20054 and CDRSB_bl>0.250 and

ICV_bl≤1697385 and RAVLT_forgetting_bl>3.500

then EMCI (0/0/169/4/1)

if Years_bl>2.471 and FAQ_bl>0.500 and SITE≤55

then LMCI (6/0/6/121/0)

if CDRSB_bl≤0.750 and ICV_bl>1489935 and

Years_bl>0.990 and RAVLT_immediate_bl>41.500

then CN (0/137/0/0/14)

if Hippocampus_bl≤7242.500 and FAQ_bl≤4.500

and Entorhinal_bl≤3037 and Fusiform_bl≤16003.500

then LMCI (0/0/0/66/0)

if MMSE_bl≤26.500 and FAQ_bl>4.500 then AD

(285 / 0 / 0 / 25 / 0)

fif CDRSB_bl≤0.750 and ICV_bl>1562475 and

WholeBrain_bl≤1121605 and PTGENDER = Male

then CN (0/106/0/1/0)

if CDRSB_bl>0.750 and ADAS11_bl≤8.500 then

LMCI (0/0/3/42/0)

if ADAS13_bl≤13.500 and SITE≤34.500 and

MidTemp_bl>20055.500 then SMC (0/5/0/0/48)

if ADAS13_bl>13.500 and

RAVLT_learning_bl>4.500 and FAQ_bl >0.500 then

EMCI (0/0/40/0/0)

if ADAS11_bl>12.500 then LMCI (0/3/0/37/0)

if Ventricles_bl≤28863.500 and MMSE_bl>28.500

and AGE>70.600 then CN (0/56/0/0/3)

if Fusiform_bl>18434 and Fusiform_bl≤18888.500

then EMCI (0/0/36/0/2)

if FAQ_bl>1.500 and MidTemp_bl≤19842 then

LMCI (0/0/0/36/0)

if MidTemp_bl≤19923.500 and

WholeBrain_bl>977838 then SMC (0/1/0/0/41)

if Ventricles_bl≤8152 then CN (0/19/0/0/2)

if PTEDUCAT≤14.500 and ADAS13_bl>17.500

then EMCI (0/0/24/0/0)

if Ventricles_bl≤30608 and CDRSB_bl>0.250 then

LMCI (0/0/0/32/0)

if AGE≤70.250 then SMC (0/1/0/0/17)

if ADAS13_bl≤11.500 and Ventricles_bl>40723

then CN (0/17/0/0/0)

if Hippocampus_bl>8041 then EMCI (0/0/13/0/0)

if ADAS13_bl>17.500 then AD (10/0/0/0/0)

if AGE≤78 then SMC (0/0/0/0/8)

else LMCI (0/0/0/1/0)

Classiﬁcation of Alzheimer’s Disease using Machine Learning Techniques

303