• No results found



Academic year: 2020



Loading.... (view fulltext now)

Full text


Abstr analy inform preva progr chall (ML) minu epide epide class corre Keyw healt of in life. intel mech being be u in th seve judg to th price Heal univ seco prob and d has notab suici tasks effec adva contr mu


Research Sch L

ract- Health car ysis, electronic mation mining ailing data, they ression muddle enges have pro ) is an applicat us being explici emiological dat emiological dat ifier machine le ect timely treatm

words - Machine

Major Depr th syndrome d nterest in activ Machine lea ligence (AI) hanically acq g explicitly pr utilized to eval he usage of s rity of the M ments of choo he patients re e saving to the

Depression i lth Organizat ersally under nd in position bable to be the developing na an assessed l ble effects o ide likelihood

Depression i s for clinician ct well-timed ancement of m

rolling of this

In the M ch research




Monal holar: Comput LDRP-ITR, K Gandhina

re area is rising patient record g that assists ba

y can realize ne es medical man ovoked the incr tion of artificial tly programme tasets. The fol tasets in terms earning algorith ment can be pro

e Learning, Ma


ressive Disord defined by ins vities, creatin arning (ML)

that offers quire and im rogrammed. M luate the depr self-reports do MDD is reco osing the type esulting in the

e patients.

is a severe he tion approac r the influenc ns of global di e figure one he ations by 2020 life span occu on class of

, and healthca

is an intricate ns regarding d cure. Thes many machine


Major Depres has been ca


ational Jou




H. Jain ter Engineerin KSV Universit agar, India

g immensely fro ds, curative tria acking for cost ew, beneficial an nagement. It ca

rease of many l intelligence (A d. Far-reaching llowing propos of its severity. hms are practice ovided which w ajor Depressive


der (MDD) i sistently depre

g main impai is an applic structures mprove from Machine learn ression epidem one by the p ognized whic e of specificcu e initial accur

ealth syndrom ching 350 m ce of depress isability probl ealth worry in

0 [1]. Major d urrence of ab life, comorbi are job.

e medical thin equally preci se tasks hav e learning tec

sive Disorde arried out a

olume 9, N

urnal of Ad


vailable On



ng Department ty

om last few dec als etc. Huge t-savings and v nd possibly life an pretense dare

machine learni AI) that offers g data mining p sed research il . Here, mean v ed to know the will result in cost Disorder, sever

is a psycholo essed mood o irment in ever cation of arti the capabilit experience m ning algorithm

miological dat atients so tha ch lets us to

ure to be deliv rate treatmen

me, with the W million indivi

sion, also stan lem and depre

equally devel depressive dis bout 17%, and

id health dis

ng that can pro se analysis an ve aggravated

chniques to ai

er (MDD) ill fter the MD

No. 2, Marc

dvanced Re


nline at ww



t A

ades. It creates size records ar verdict constru e-saving inform es for clinician ing techniques

structures the c practices utilizin lustrates the id value substitutio severity of the t saving and fas rity, data prepro

ogical r loss ryday ificial ty to minus ms can tasets at the take vered nt and World iduals nding ession loped sorder d has sease, oduce nd in d the id the ness, DD is ide nec the gro mu pha gro lear prim pre wh trea healt loss day-along furth prob unip encu deve natio viole MDD searc Afte an en

ch-April 20

esearch in C





Assistant Prof L

large amount o re a key sourc ucting. When m mation. Heteroge ns concerning c to aid advance capability to m ng machine lear dea of machin on is performed

depression am ster improveme ocessing, multi ntified and cessity to pro

outcomes ounds of fMR

ch research ase of MDD ounded on rning techni mary stage. A edict the sev

o benefits th atment to get


Major Depre th syndrome of interest in to-day life. g with the hermore event Major Depre blem. Agreein olar depressi umbrance list eloped nations ons. Depressio ence, strokes, D is frequentl ch for aid wh

r recovering u nd to cure unt






Mehul fessor: Compu

LDRP-ITR, K Gandhina

of records about ce which needs medical organiz eneity of major correct analysis e the controlling mechanically acq rning procedure ne learning alg

d on the data t mong the people ent of health con

class classificat originates ovide the cu are examin RI done on has been ca where speci the self-rep iques can le

Also because verity of MD he specific to t cured early


essive Disord described by n activities, p MDD is a s suffering im tually can lead

essive Disor ng to the Wor

ion is on p t of malady. s and first in d

on orders upp , and transpo ly intermitten hile in the m up to their lev til the cycle be





P. Barot uter Engineeri KSV Universit agar,India

t patients, clinic s preprocessing zations apply d depressive diso s and actual we

g of this diseas quire and impr es have potenti orithms to ana to preprocess it e at an early stag

nditions of pati tion.

severe wh ure to the spe ned and ant the enduring arried out on ific reports h ports by ap t us to pred e of this one DD existence o take precau y.


der (MDD) i persistently d producing ma severe psych mmense sens d to demise by

rder (MDD) rld Health Or

osition three It orders e developing and

per than ische ortation accide nt and lasting, middle of a de el of function egins another

o. 0976-5697


ing Departmen ty

cal sources, inf g and evaluatio data mining on order (MDD) d ell-timed cure. se. Machine lea

ove from exper al in the study alyze the depre t and then mult ge so that prope ents.

hich creates ecific. Gener ticipated on g individual. n the prelimi himself. Rese pplying mac dict the MD

can evaluate e in the spe utions and ti


s a psycholo depressed mo ajor impairme

ological state itive agony, y suicide.

is a world rganization (2 globally on ighth among d highly devel emic heart ill ents. Even th persons may epressive inci nality, they ma


Monal H. Jainet al, International Journal of Advanced Research in Computer Science, 9 (2), March-April 2018,46-52

© 2015-19, IJARCS All Rights Reserved 47

The DSM-IV TR (American Psychological Association, 2000) finds and talks over the analytical conditions. To have the conditions for MDD, minimum five of the signs enumerated beneath need to exist in the similar 2-week duration, and must characterize a transformation in the distinct. The signs should also cause major sorrow or loss in public, profession, or other fields [2]. The signs shouldn’t be the reason of utilization of any kind of stuff, a health state, another psychological syndrome, or grief [2]. Also, the specific can’t havehyper in the past, hypomanic, or mixed incidence [2]. It is furthermore essential to annotate that those with MDD are at a greater than before threat for evolving co-morbid mental disorder such as nervousness, frights, whim control, or substance misuse.


Onset of a Major Depressive Episode can be anywhere from days to a few weeks. Symptoms must be present for most of the day, nearly every day, for at least 2-weeks and include:

 Depressed Mood

 Diminished interest/pleasure

 Significant weight loss or gain

 Insomnia or hypersomnia

 Psychomotor agitation or retardation

 Fatigue

 Feelings of worthlessness or guilt

 Difficulty concentrating

 Recurrent thoughts of death/suicidal ideation

Depressed mood: Depressed mood is considered as the promising indicator of MDD. It is defined as sense ofunhappiness, downhearted, or discouraged. Few individuals don’t have the ability to describe their emotions while other few individuals don’t want to agree of sensing depression.Few individuals might agree of having bad temper along with or instead of having depressed mood. It is significant to note the effect of the person, giving close consideration to the reactions on face, stance, and tendency of speech. This is specifically significant if the individual is in rejection about his or her mental state.

Loss of interest in activities: Depressed persons will frequently observe that they are not finding things interesting which they used to enjoy earlier. Some call it as not viewing frontward to things, or being incapable to feel pleasure. It might come into notice of some people that the affected individual is isolating himself from his family and friends. A loss of sexual desire may also be noticed.

Weight changes: Hungriness varies bring about in noteworthy, unplanned variation in weight are a lot observed in MDD. This might be seen as craving for fat and carbohydrate foods in some beings while appetite loss in other beings.

Sleep changes: Sleeplessness is mostly observed in MDD. Beings may discover themselves awake in the midnight and are not capable to again asleep. Others may lie wakeful, unable to initiate forty winks. In some occurrences, hypersomnia happens.

Psychomotor changes: Psychomotor changes should be observable by others and take account of agitation causing in walking, leg rebounding, or twitching. Also, retardation may mark as speaking and moving slowly or delay in reacting to questions.

Fatigue: Too much fatigue is often a indication that really influences the subject. He or she may have non existence of the liveliness to do the day-to-day chores of existing. Weariness, even after a wholedusk’snap, is common.

Feelings of worthlessness or excessive or inappropriate guilt: Those be subjected to a major depressive episode repeatedly dock strong feelings of insignificance and culpability. Individual may sense worthless of the belongings in their existence. They may preoccupy and know-how strong culpability over former or existent happenings. Individual also might adversely misjudge things spoken or done by others. This continues the culpability and feelings of contemptibility.

Indecisiveness and concentration problems: Individuals having MDD repeatedly experience trouble focusing on responsibilities. This must be an alteration from usual working. Concentrating problems also ascend, and can consequence in professional, scholastic, and association difficulties. Creating choices can unexpectedly seem mammoth to a blue individual.

Recurrent thoughts of death and/or suicide: The major fear with Major Depressive Disorder is that of suicide. Considerations of death are a common incidence in the distressed. These views may differ liable on the sternness of the depression. It is further vis-à-vis if the person has made a strategy of how he or she would do suicide. Regrettably, some follows the strategy and take their own breathes.

A. PHQ-9 as a Diagnostic Measure


Figure 1: PHQ-9 Symptom Checklist[3]

As a severity degree, the PHQ-9 markarrays from 0 to 27, asevery of the 9 items can be counted from 0 (“not at all”) to 3 (“nearly every day”). Easy-to-remember change points of 5, 10, 15, and 20 signify the onsets for mild, moderate, moderately severe, and severe depression, respectively [3]. Advisedcureactivities in reaction to these differentstages of PHQ-9 depression severity are shown in Table1. Scores less than 10 seldom occur in individuals with major depression whereas scores of 15 or greater usually signify the presence of major depression.


PHQ-9 Score

Depression Severity

Proposed Treatment

1 to 4 None None

5 to 9 Mild On the lookoutcoming up, PHQ-9 follow up

10 to 14 Moderate Strategize cure, Think throughtherapy 15 to 19 Moderately


Abruptbeginning of pharmacotherapy/psychothe rapy

20 to 27 Severe If severe diminishing or lowlyreaction to

psychotherapy, talk about to psychological health professional for psychiatric therapy and



One of the subareas of artificial intelligence (AI) is Machine Learning[4]. The objective of machine learning commonly is to realize the structure of data and fit that data into models that can be understood and exploited by people. Machine learning systems in its place let computers to train on data inputs and practice statistical analysis in order to output values that fall inside a particular range. As of this, machine learning enables computers in building models from sample data in order to mechanize decision-making procedures centered on data inputs. Two of the most broadly accepted machine learning approaches is supervised learning which trains algorithms based on example input and output data that is labeled by humans, and unsupervised learning which provides the algorithm with no labeled data in order to allow it to find structure within its input data.

Supervised Learning

In supervised learning, the computer is delivered with sample inputs that are characterized with their preferred outputs. The determination of this technique is for the algorithm to be capable to “learn” by equaling its real output with the “taught” outputs to discover errors, and transform the model accordingly. Supervised learning therefore uses patterns to guess label values on additional unlabeled data. A common use case of supervised learning is to utilize past data to predict statistically probable future happenings.

Unsupervised Learning

In unsupervised learning, data is unlabeled, so the learning algorithm is left to discover commonalities among its input data. As unlabeled data are more ample than labeled data, machine learning methods that assist unsupervised learning are principally valuable. The objective of unsupervised learning may be as direct as determining hidden patterns inside a dataset, but it may also have a objective of feature learning, which lets the computational machine to mechanically learn the representations that are required to categorize raw data.

Learning approaches can be classified into linear and nonlinear approaches. Linear approaches are simpler, while nonlinear approaches are more flexible in behavior. For supervised learning, the methods can be additionally classified as classification- or regression-based methods. Classification-based methods try to classify the data by discrete and categorical labels, while regression-based methods fit the data to a continuous function and thus work with continuous labels for the data [5]. For unsupervised learning, the methods are primarily categorized as clustering method, which group the data into clusters based on underlying similarities [5].


In the current works,a nature motivated and unique FS algorithm grounded on standard Ant Colony Optimization(ACO), known as improved ACO(IACO), was utilized to lessen the number of attributes by eliminating unrelated and out of work information [6]. The nominated attributes were then providing for into support vector machine (SVM), in order to categorize MDD and BD subjects [6]. Quantitative electroencephalography (QEEG) coherence data of 46 BD and 55 MDD subjects were fed into IACO first [6]. Selected more informative feature subset from 16 electrodes of alpha, delta and theta frequency bands was then used as input in SVM [6]. Enactment of IACO–SVM methodology identified that it is promising to distinguish 46 BD 55 MDD themes using 22 of 48 attributes with 80.19% total taxonomy accuracy (Turker Tekin Erguzel et al., 2015).

Current works also determine that guilt-selective modifications in functional connectivity of the ATL are satisfactory to differentiate the remitted MD set from the control set with in elevation accuracy [7]. The outcomes demonstrate that guilt-selective functional disconnection of the ATL has the impending to be moreestablished into a clinically beneficial fMRI biomarker of MD liability [7]. Expending a lately recognized neural signature of guilt-selective functional disconnection, the machine learning classification procedure was capable to differentiate remitted MD from control members with 78.3% accuracy [7]. This determines the high possibility of our fMRI signature as a biomarker of MD vulnerability (João R. Sato et al., 2015).


Monal H. Jainet al, International Journal of Advanced Research in Computer Science, 9 (2), March-April 2018,46-52

© 2015-19, IJARCS All Rights Reserved 49

logical psychopathology based interview (AMDP) including novel items to assess moral emotions (n=94 patients) and the structured clinical interview-I for DSM-IV-TR [8]. Cluster analysis was engaged to recognize indicator rationality for the most severe incidence.As anticipated, moods of insufficiency and uselessness were part of the central depressive disorder, closely co-occurring with depressed frame of mind. Self-blaming feelings were very recurrent and concerning but not limited to guilt [8]. This calls for a sophisticated valuation of self-blaming sentiments to progress the identification and stratification of MDD (Roland Zahn et al., 2015).

Current works also utilized a three-step procedure fusing multiple imputation, a machine learning boosted regression procedure and logistic regression, to find key biomarkers related with depression in the NHANES (2009– 2010) [1]. After the formation of 20 imputation data collections from multiple chained regression sequences, machine learning boosted regression at first identified 21 biomarkers related with depression [1]. Using traditional logistic regression methods, countingmonitoring for probable confounders and moderators, an ultimate set of three biomarkers were nominated [1]. The methodical utilization of a amalgam procedure for variable choice, merging data mining procedures consuming a machine learning procedure with traditional statistical modeling, accounted for lost information and composite survey selection approach and was validated to be a suitable tool for spotting three biomarkers related with depression for future theory generation: red cell distribution width, serum glucose and total bilirubin (Joanna F. Dipnall et al., 2016).

Also, the WMH ML prototypes were practiced to this baseline information to produce expected result scores that were matched to detected marks measured 10–12 years after baseline [9]. ML prototypical expectation accuracy was also equaled to that of conventional logistic regression models [9]. Area under the receiver operating characteristic curve (AUC) centered on ML (.63 for high chronicity and .71–.76 for the other probable outcomes) was dependably greater than for the logistic models (.62–.70) even though the later models with more predictors [9]. 34.6–38.1% of individuals with consequent high persistence-chronicity and 40.8–55.8% with the severity signs were in the top 20% of the baseline ML predicted risk distribution, while only 0.9% of respondents with consequent hospitalizations and 1.5% with suicide trieswere in the lowermost 20% of the ML predicted risk distribution [9]. Outcomes approve that clinically beneficial MDD risk stratification representations can be produced from baseline patient self-reports and that ML approaches advance on conventional procedures in emerging such models (Ronald C. Kessler et al., 2016).


A. Study participants and measurements

In this research study two different datasets were utilized. One dataset is collected from the National Health and Nutrition Examination Survey (NHANES) (2015–2016) were utilized. Relevant NHANES data files were downloaded from the website. NHANES is a cross-sectional, population-based study of civilians aged 18 to 80 years, conducted in two-year blocks. The sample size of this dataset is 5735[10].

Second dataset is collected from LDRP-ITR college of KSV University, Gandhinagar(INDIA). The respondents aged 16 to 25 years in the study. The sample size of this dataset is 342.

The Patient Health Questonnaire-9 (PHQ-9) was cast-off to evaluate depressive indicators(depression). The PHQ-9 is a well-validated, self-report instrument for discovering and observing depression, with worthy concordance with a medical analysis of major depressive disorder (MDD). Things evaluate the existence of nine Diagnostic and Statistical Manual of Mental Disorder Fourth Edition (DSM-IV) depression indicators over the past two weeks, and are counted on a four-point scale representing the grade of severity from 0 (not at all) to 3 (nearlyevery day). Things were then totaled to form a complete severity mark going from 0 to 27 where those with a over-all score of 10 or more were measured depressed (i.e. moderately to severely depressed).

B. Methodology

The two datasets collected for the research study contains missing values, which needs to be processed after which the ML classifiers are implemented to predict the severity of MDD.

Data Preprocessing

In several ‘big data’ circumstances, absent information are not a subject due to the large size of observations and variables or features existing. In difference, absent data in readings with lesser model amounts can influence the results greatly. There have been a several missing data techniques recommended in the works over the last years: listwise deletion; pairwise deletion; mean substitution; regression imputation; Maximum Likelihood (ML) estimation. Most of these techniques can only be cast-off when there is no arrangement for the missing data. The selection of techniquecast-off for handling with missing data is frequently not as much of significant when the ratio of missing data is less than 5%. Besides, it is not rare for the ratio of missing data in big epidemiological readings to surpass this fraction.

Mean substitution is a common strategy for addressing the missing value problem. In the research, attribute mean value is used to replace all the missing values in both the datasets.

ML classifiers

Machine learning classifiers are implemented to predict the severity of MDD present among respondents. MDD is assessed using PHQ-9. PHQ-9 measures the severity based on the 0-27 score as none (0-4 score), mild (5-9 score), moderate (10-14 score), moderately severe (15-19) and severe (20-27 score). Machine learning multiclass classifiers can be implemented to predict the type of severity of MDD present among respondents. Several multiclass classifiers such as Support Vector Machine (SVM) classifier, Decision Tree, k-Nearest Neighbors, Naïve Bayes classifier can be implemented to classify the severity of MDD. These classifiers can be cast-off for multiclass classification issue by extending the binary classification methods.


effec upon mini close binar hand these optim class Two trees [11]. infor creat cente infor [11]. the r impl node The class the K class insta Eucl is ev and delib norm valid

the n prob prob to an that with expe expr P(M As th from class avail expla Support V ctive classific n the notion o imum space estsample [1

ry classificati dle the multic

e additions, e mization prob ses [11].


Decision o broadly reco

s are Classifi . The tree a rmation create te a good sim ered on the rmation gain [ . A different i root nodule to lemented on c e reached is c

algorithm c sification prob K classes conc

k- Neare

kNN is c sification alg ance, the exp lidean) from th valuated [11].

the most c berated the o mally determi

dation [11].

Naïve Ba

Naive Ba norm of Max blem with K babilities P(M1

n unidentified m = argmaxm h the maxim erimental data ressed, utili =m||a1,...,aN)= he denominat m the evaluati s conditional

lable classes [ anation the de

Vector Machi cation proced of exploiting from the 1]. The elem on, but additi class classific extra factors a blem to hand

n Tree

n trees are a po ognized proce ication and R attempts to d ed on the valu mplification [1 e attribute [11]. Every le nstance is cla a leaf nodule certain attribut considered th can naturally

blems. The le cerned [11].

est Neighbors

considered am gorithms [11]

panse (using hat instance to . The k mini characterized output class l

ined using a


ayes is an ef ximum A Post

classes {M1, 1), . . . , P(MK instance with mP(M =m||a1, mum a poste

a [11]. This izing Baye = P(M=m)P(a

tor is the same ion. Now, we

probabilities [11]. This can ependencies a

ines are amid t dures [11]. T the margin i. separating h mentary SVM

ions have been cation instanc

and restraints le the separa

owerful classi edures for con Regression Tr

deduce a spl ues of the acce

1]. The split that provide eaf nodule rela ssified by foll e, where at eve

te of that insta he class label handle bin eaf nodes can


mong the olde . To classify some dista o every furthe mum distanc

class in th label [11]. T validation se

ffective classif teriori (MAP) . . . ,MK} w K), we can allo h attributes a = . . . , aN), that erior probabi aposterior p es theorem a1,...,aN||M=m e for all class e should calcu

of the attrib be quite chal mong attribut

the best robus They are cen e. maximizin yperplane to M provisions

n recommend e as well [11 s are added t ation of the u

ification techn nstructing dec rees and ID3/ it of the tra essible attribu at every nodu es the maxi

ates to a class lowing a path ery nodule a t ance [11]. The

for that exam ary or multi n refer to eith

est non-param fy an uniden ance measure er training ins es are recogn hese k classe The value of et or using c

fier centered ) [11]. Assum with so-called ot the class lab = (a1, . . . , aN)

t is select the ility assumed probability ca m, as fol m)/P(a1,...,aN)

es, it can be l ulate the so-c butes assumed

lenging taking tes [11]. The N

st and ntered ng the o the only ded to 1]. In to the unlike nique. cision /C4.5 aining utes to ule is imum label from test is e leaf mple. iclass her of metric ntified e e.g. stance nized, es is k is cross-upon med a prior bel m ) such class d the an be lows: [11]. let go called d the g into Naive Baye indep class = m maxi Evid havin well cond mult indiv WEK Ana writt Zeal proc eithe Java class visua mach suita datas MDD WEK on th are s NHA Logi class in f insta Deci preci follo Whil insta and D F 10 20 30 40 50 60 70 es methodol pendence i.e. s [11]. This sh m)… P(aN||M

imizes this v dently this me ng more than inspite of ditional indepe The per ticlassifiers to viduals areas KA.Waikato

lysis (Weka) ten in Java, es and [12]. W edures for da er be practiced a code[13]. W

sification, reg alization [13] hine learning able ML mult sets to obtain D among resp

KA tool is util he two dataset shown below.

ANES datase

istic, SVM, k s classifiers ar figure, Logis ances (records ision Table. ision, recall, owed by SVM le Logistic Cl ances and FP Decision Tabl

igure 2: Graph of 0 000 000 000 000 000 000 000 Logistic Correctly

logy is to x1, . . . , xN hortens the nu = m), and th value over all

thod is certain two classes, a the origina endence [11].

rformance of o predict the

ssessed by


is a collectio stablished at t Weka is a g ata mining err d straight to a Weka comprise gression, clus ]. It is also g patterns. Fr ticlass classif

better results pondents.


lized to imple ts to predict th


-NN, Naïve B re applied to t stic classifier s) followed by Also, Logisti F-measure, M, k-NN, Naï

lassifier have rate followed le. The results

f correctly and in class

c SVM k

y classified

o assume c N are indepen umerator to be hen selecting l the classes nly extensible and was prese al simplifying

f the above severity of utilizing D

nment fo

on of machine the University gathering of rands [13]. T

dataset or cal es tools for da

stering, assoc well-suited rom the resu fier can be im s of classifica


ment ML mul he severity of

Bayes and De the dataset in r classified y SVM, k-NN ic results hav

TP rate and ïve Bayes an

the lowest inc d by SVM, k-s are k-shown in

correctly classifie sifiers k‐NN Decisio Table Incorrectly class condit ndent specifie e P(M = m)P( g the class m m = 1, . . e to the instan ented to imple g suppositio

e mentioned MDD among Data mining

or Knowl

e learning soft y of Waikato,

machine lea he procedure lled from your

ata pre-proces ciation rules, for evolvingn ults of Weka mplemented o ation of severi

lticlass classif the MDD. Re

ecision Table WEKA. As sh the most co N, Naïve Baye

ve high accu d Kappa Sta nd Decision T correctly clas -NN, Naïve B n below graph

ed instances by M


© 201




Logi class in f insta Deci recal Baye have follo 0 1


15-19, IJARCS A

Figure 3

ure 4: Graph of pr

re 5: Graph of Ka

RP-ITR datas

istic, SVM, N s classifiers ar figure, Logis ances (records ision Table. ll, TP rate an es, k-NN and e the lowest owed by SVM

0 .5 1 .5


Logistic SV

Logistic SV

onal H. Jainet al

All Rights Reserve

: Graph of TP rat

recision, recall, F

appa Statistic, Me error of ML


Naïve Bayes, k re applied to t stic classifier s) followed by Also, Logisti nd Kappa Stat d Decision T

incorrectly cl M, Naïve Bay


TP rate



F measure


Kappa Stati

Mean absol

Root mean 

l, International J


te & FP rate ML c

F measure & accu

ean absolute error L classifiers

k-NN and De the dataset in r classified y SVM, Naïve ic results hav tistic followe able. While L lassified insta yes, k-NN an

k‐NN Decisi


FP rate

N Decision 

Table Recall 


N Decision 

Table stic

lute error

squared error

Journal of Advan


uracy of ML class

r & Root mean sq

ecision Table WEKA. As sh the most co e Bayes, k-NN ve high accu

d by SVM, N Logistic Clas ances and FP nd Decision T

ion  e

Naïve  Bayes

Naïve  Bayes

Naïve  Bayes r

nced Research in



multi hown orrect N and uracy, Naïve ssifier P rate Table.

Aver Deci insta


Figu 5 10 15 20 25 30 35

0. 0. 0. 0. 1.

n Computer Scie

rage precision ision Table a ances.The resu

igure 6: Graph of

Figure 7

ure 8: Graph of pr 0

50 00 50 00 50 00 50




0 .2 .4 .6 .8 1 .2


Logistic S

ence, 9 (2), Marc

n and F mea and Naïve Ba ults are shown

f correctly and in class

: Graph of TP rat

recision, recall, F

SVM k‐

Correctly Class

ncorrectly Cla

SVM k‐

TP rate



F measure

ch-April 2018,46

asure isn’t cal ayes as there n in below gra

correctly classifie sifiers

te & FP rate ML c

F measure & accu ‐NN Decision


sified Instance

ssified Instanc

NN Decision Table 

FP rate

NN Decision

Table  Recall

Accurac 6-52

51 lculated for S e are small n aph.

ed instances by M


uracy of ML class

n  Naïve 




n  Naïve 


n  Naïve 

Bayes y

1 SVM, no of




epide mult value the p done is no futur


re 9: Graph of Ka

Thus M emiological d ticlass classifi e substitution preprocessed e using Weka. ot considered re work.

Joanna F. D Williams, S “Fusing Da Statistics to February 5 pone.014819

Logistic SV

appa Statistic, Me error of ML


MDD severity dataset based o fiers. Here da n, after which data to predi . Also in this d which can b


Dipnall, Julie A Seetal Dodd, ata Mining, M Detect Biomar 5, 2016, PL 95.


Kappa Stati

Mean Abso

Root mean 

ean absolute error L classifiers


y can be on the baseline ata is preproc

evaluation of ct the severity

research stud be taken into


A. Pasco, Mic Felice N. Jac Machine Learni rkers Associate OS ONE, D

N Decision 

Table  stic

lute Error

squared error

r & Root mean sq

assessed on e reports using cessed using f ML classifie y of MDD ca dy, the comorb o consideratio

chael Berk, La cka, Denny M ing and Tradi ed with Depres DOI:10.1371/jo

Naïve  Bayes r


n the g ML mean ers on an be bidity on for

ana J. Meyer,

itional ssion”, ournal.



[4] [5]





[10] [11]

[12] [13]

Stacie L. N Treatment unpublished Kurt Kroen Depression Annals 32:9, https://www introduction-Meenal J. “Studying d methods”, N Turker Tekin based appro major depre Biology and Joao R. Sato Carlos E. Th accurately d depression”, pp.289-291. Roland Zah Green, John “The role psychopatho Affective Di Ronald C. K Robert M. Daniel Eber Andrew A. Rosellini, N A. Wilcox, learning alg major depre Psychiatry 2 https://www ponent=Que Mohamed Methods”, https://pdfs.s b39a3ebefda 649-9526493 https://www https://en.wi

Nelson, “Major and Recurren .

nke, Robert L Diagnostic an , 2002.

.digitalocean.co -to-machine-lea Patel, Alexand depression usin NeuroImage: Cli n Erguzel, Cum oach for featu

ssive disorder Medicine 64(2 o, Jorge Moll, homaz, Roland detects fMRI s Psychiatry Re hn, Karen E. L

F. William De of self bla ology of major isorders 186 (20 Kessler, Hanna Bossarte, Lisa rt, Irving Hwa Nierenberg, Nancy A. Samp EdD, Alan M orithm to pred essive disorder

016 October 21 n.cdc.gov/nchs stionnaire&Cyc

Aly, “Survey semanticscholar ae694.pdf?_ga= 322.151934764 .cs.waikato.ac.n ikipedia.org/wik

r Depressive D nce Preventio

. Spitzer, “Th nd Severity M om/community arning. der Khalaf, Ho ng imaging an

inical 10(2016) mhur Tas, Merv ure selection a

– bipolar disor 2015) , pp.127-1 Sophie Green, Zahn, “Machin signature of vu esearch: Neuroi Lythe, Jennifer eakin, Allan H. ame and wo r depressive d 015), pp. 337-3 M. van Loo, a A. Brenner, ang, Junlong L Maria V. Petu pson, Robert A M. Zaslavsky, dict the persiste from baseline 1(10): 1366-137 s/nhanes/search/ cleBeginYear= y on Multic

November r.org/a546/f2c8 =2.146174396.9 49.

nz/ml/weka/ ki/Weka_(mach

Disorder: Over on”, August he PHQ-9: A Measure”, Psych

/tutorials/an-oward J. Aizen nd machine lea ), pp. 115-123.

ve Cebi, “A wr and classificati

rders”, Comput 137.

, John F.W. D ne learning algo ulnerability to

maging 233 (2 r A. Gethin, S Young, Jorge orthlessness in disorder”, Journ


Klaas J. Ward Tianxi Cai, D Li, Peter de J ukhova, Antho A. Schoevers, M “Testing a ma ence and sever e self reports” 71.

/datapage.aspx? 2015

class Classifi 88c588a2a46c0 963036132.151

hine_learning) rview,

2012, New hiatric

nstein, arning rapper on of ters in

eakin, orithm major 015) , Sophie Moll, n the

nal of denaar,

David Jonge, ony J. Marsha

achine rity of

, Mol ?Com cation


Figure 1: PHQ-9 Symptom Checklist[3]


Related documents

The preferential localization of mutant htt in the nucleus of striatal neurons suggests that the nuclear effect of mutant htt or gene transcriptional dysregulation may affect neuro-

This basically deals with the queries of user for a company simultaneously running its e-courier service where the users having booked some couriers or both for delivery can

 trading  partners   currently  participating  in  those

P -value of < 0.2 in univariate analysis were entered in multivariable logistic analysis with delayed initiation of breastfeeding as the dependent variable and the other

Addition of 5%, 10% and 15% of JRP in wheat flour caused significantly (p < 0.05) higher insoluble, soluble and total dietary fiber in flour and bread products.. Results

Keywords: Fermented Beverage, Gram Positive, Health Promoting, Probiotic Bacterial Isolates, Starch Test..

A lot of data modeling stud- ies has already been conducted by different researchers for NoSQL databases, but non has provided in depth study for the ANT + sensor data especially

Institutional review board approval was obtained and a ret- rospective chart search was conducted to identify all patients from January 2003 to October 2013 with a diagnosis