Neural Approach for Determining Mental Health Problems

Published on December 2016 | Categories: Documents | Downloads: 54 | Comments: 0 | Views: 286
of 6
Download PDF   Embed   Report

Journal of Computing, eISSN 2151-9617, http://www.JournalofComputing.org

Comments

Content


Neural Approach for Determining Mental
Health Problems
Jabar H.Yousif, Mabruk A. Fekihal
Abstract—Mental illness is to become one of the main problems in our society. World Health Organization predicted that
depression will be the main cause of the world's leading disability by 2020. This paper aims to implement a soft computing
technique to determine mental health problems. A multilayered perceptron designed and implemented to classify transcribed
speech samples and determine a mental health problem. The NeuroSolution package is used to adopt the neural network
phase. The Error backpropagation learning techniques is implemented to see how effective they would be at correctly predicting
the classification of text. The proposed classification system is used to determine if a text or speech sample was generated by a
person has a mental health problem such as schizophrenia or mania. Classifications such as schizophrenia are very broad;
therefore, the approach introduced here arrives at practical and task-relevant diagnostic categories by use of clustering
techniques. The results demonstrate high accuracy (99%).
Index Terms— Mental Illness, Text clustering, Text Classification, Soft Computing, Neural Networks.
——————————

——————————
1 INTRODUCTION
Due to the rapid development in information technology
and significant growth in data transfer, therefore extensive
studies are made to achieve rapid progress in mining, clas-
sification and extraction of data [9,10]. As a result, over the
past years, a number of approaches are developed for au-
tomatic text classification of the data [19]. Aim of most
researches in the field of text classification is improving the
speed and accuracy of data classification [4,13].
Researchers are increasingly looking forward to the dis-
covery of new techniques and innovative by using of in-
formation technology that helps to overcome the rapid rise
in health care costs faced by the community. Research has
shown that in the past held that the use of artificial intelli-
gence techniques (AI) can help diagnose diseases and as-
sess treatment outcomes [21,23].
Soft computing (SC) refers to methods of calculation
that assembly assay to discover the approximate solutions
for the formulation of satisfying real-world problems. It
aims to build a tool to have human capacities such as
thinking, learning and thinking, problem solving, etc. The
recent developments in the field of information technology
and the use of soft computing techniques, particularly
neural networks and fuzzy logic and support vector ma-
chines can help to develop effective automated diagnostic
systems. In spite of the great challenges in the future, the
use of new developments in the field of artificial intelli-
gence techniques promise a lot in solving the problems of
medical and health related.
The artificial neural network is one of the most qualified
techniques for the learning from the rare data. The purpose
of this paper is to design and implement a text classifica-
tion for mental health based multilayer perceptron neural
network. There are specific features that make neural net-
works successfully applied in a number of applications[3].
Professional psychiatrists work hard consciously and
unconsciously to investigate the language of their patients
with the intention of identifying patterns, and using these
patterns to assist in building an obvious clinical diagnosis
[24].
Clustering is a separation of data into groups of similar
objects. Clustering is one of the important topics of re-
search that proved the important successes in many fields
such as statistics, pattern recognition and machine learning
[11].
In the process of text classification, the majority of ma-
chine learning algorithms require labeled samples for train-
ing. Nevertheless, the obtaining of labeled samples is very
limited [6].
2 MENTAL HEALTH BACKGROUND
The term "mental health" generally refers to a psychologi-
cal and emotional state. Mental illness is to become one of
the main problems in our society [2]. World Health Or-
ganization predicted that depression would be the main
cause of the world's leading disability by 2020. Research
in the field of mental health has increased and resulted in
extensive collection of information and publications cov-
ering different features of mental health and utilizing a
wide range of problems [7,8,16].
Mental health care services are commuting its focal
point from inpatient to community care. Community based
psychosocial treatment program are broadly advocated to
present wide-ranging care to peoples who have a mental
health illness [17,18]. It is extremely frequent and cause
substantial social and economic onus worldwide. Never-
————————————————
- Jabar H. Yousif, Faculty of Computing and Information Technology, Sohar
University, Oman.
- Mabruk. A. Fekihal, Faculty of Computing and Information Technology,
Sohar University, Oman.


2012 Journal of Computing Press, NY, USA, ISSN 2151-9617
http://sites.google.com/site/journalofcomputing/
JOURNAL OF COMPUTING, VOLUME 4, ISSUE 1, JANUARY 2012, ISSN 2151-9617
https://sites.google.com/site/journalofcomputing
WWW.JOURNALOFCOMPUTING.ORG 6

theless, no standardized biological diagnostic tests are
available, and diagnosis is still dependent on clinical skills
and opinion.
Mental health illness can increase at any position in life-
time which possibly influenced by a number of factors
such as genetics or family history of a neuropathy , the
pressures of work and the difficulties of life present[17].
There are no yet any diagnostic tests, biological, and still
the diagnosis depends on clinical experience skills of the
physician and his opinion. There are a number of efforts to
measure the clinical diagnosis, particularly through the
development of questionnaires and diagnostic interviews
and systematic examination, besides, a number of re-
searches in direction of examine the movements of the face
in certain disorders [8,10,11].
In recent years there have been an extensive attempts
made to develop potential biological or genetic markets,
but it looks obvious that plain biological diagnostic tests for
psychiatric disorders are still a few years away.

3 NEURAL NETWORKS BACKGROUND
A neural network is a powerful data modeling tool that is
able to capture and represent complex input/output rela-
tionships either linear and non-linear one[12] .There are
specific features that stimulate scientists to adopt neural
network design theme in the different application fields.
The main features are Massive parallelism, Uniformity,
distribution representation and computation, learning
ability, Trainability, generalization ability, and adaptivity
[14, 15]. Neural networks are being successfully applied
in a number of areas such as data classifications, resource
allocation and scheduling, database mining, speech pro-
duction and recognition and pattern recognition [12,23].
The multilayer perceptron (MLP) is considered as one
of the most common neural network models. The MLP
neural network is one type of a supervised network be-
cause it requires a desired output in order to learn. The
main objectives of this type of network are to engender a
model that correctly maps the input to the output using
previous knowledge and it is perform the tagging task
with low process time [15].
4 RELATED WORK
Joachims [10,11] introduces a transductive method for
text classification using support vector machines and ob-
tained significant improvements in the classification task.
He achieved in the classification phase accuracy of 80%
for the schizophrenia versus control task.
Kiritchenko [22] implements co-training method to cate-
gorize the email with unlabeled samples. The experi-
ments with co-training based on the email domain are
implemented. The results show that the performance of
co-training depends on the used learning algorithm. In
particular, Support Vector Machines achieved 90.06% and
Naive Bayes achieved 80.36% on email classification.
Mabruk and Joachim [1] they used text mining technolo-
gy to determine psychiatric problem. They examined if a
text or speech sample is generated by a person with a
mental health problem such as schizophrenia or mania.
The results show very good accuracy (92%) and precision
(87%) and moderate to low recall.
5 MLP CONFIGURATION & DESIGN
The multilayer perceptron (MLP) is one of the most wide-
ly implemented neural network topologies. The discrimi-
nant functions can take any shape, as required by the in-
put data clusters. Moreover, when the weights are prop-
erly normalized and the output classes are normalized to
0/1, the MLP achieves the performance of the maximum
a posteriori receiver, which is optimal from a classifica-
tion point of view [12,15]. A MLP network with error
back-propagation learning algorithm is used [20]. The
backpropagation rule propagates the errors through the
network and allows adaptation of the hidden PEs. The
error correction learning works is the response at PE i at
iteration n, yi(n), and the desired response di(n) for a giv-
en input pattern an instantaneous error ei (n) is defined
by :

) 1 )...( ( (n) d (n) e
i i
n y
i
÷ =

The theory of gradient descent learning is used to adapt
each weight in the network by correcting the present value
of the weight as follows:

) 2 )...( ( ) ( (n) w 1) (n w
ij ij
n x n
j i
+ + = + qo

Where the local error
) ( n
i
o
is computed from ei (n) at
the output PE or can be computed as a weighted sum of
errors at the internal PEs. The constant step size isq .
Momentum learning is an improvement to the straight
gradient descent in the sense that a memory term (the past
increment to the weight) is used to speed up and stabilize
convergence. The momentum learning is used to update
the weights of nerwork as illustrated in equation 3:

+ + + = + ) ( ) ( (n) w 1) (n w
ij ij
n x n
j i
qo
...(3) 1)) - (n w - (n) w (
ij ij
o

The
o
is the momentum value. Normally is to be set
between 0.1 and 0.9, the best value is 0.7.
In this paper, we used a multilayer perceptron (MLP) as
illustrated in Figure.1. Our network has one hidden layer,
11 Processing Element (PEs) as input (I1,I2,..,I11) , while it
has 6 Processing Element as output (O1,O2,…,O6).











JOURNAL OF COMPUTING, VOLUME 4, ISSUE 1, JANUARY 2012, ISSN 2151-9617
https://sites.google.com/site/journalofcomputing
WWW.JOURNALOFCOMPUTING.ORG 7

FIGURE.1 MLP WITH ONE HIDDEN LAYER


The maximum number of epochs is 1000. The TanhAx-
on transfer function is implemented in hidden layer, as
well it is implemented in output layer. The TanhAxon ap-
plies a bias and Tanh function over each neuron in the
layer. This will squash the range of each neuron in the
layer to value between (-1) and (1). However, in sake of
giving the network the ability to construct soft decisions,
the nonlinear elements are used.
The momentum learning rule is adopted with constant
step size equal to (1), and momentum rate equal to (0.7).
6 EXPERIMENTS AND RESULTS
Two hundred messages from each "sci.psychology psy-
chotherapy" and "sci.philosophy" are selected randomly
and made unidentified by deleting header and signature
information [1,10]. The two newsgroups are chosen ac-
cording to the previous discusses ideas in psychotherapy
including diagnosis and the later may include similar
content but generally must be dissimilar. The classifica-
tion mission is difficult since the topic area of the two
newsgroups is so similar that cross-posting can occur.
The experiments are performed using 200 text files in-
clude the original messages inaddition to their replies.
Further, a pre-processing is implemented on the text
samples in order to encoding them into a binary form
which can be used by the network. Therefore, Inputs are
transcribed speech samples in binary form. And, the out-
put will determine whether the text has been written by
someone who has mental problems by searching for key
symptoms of mental illness in the text such as attention,
deficit, hyperactivity, and disorder (ADHD) [1,10,11].
The experiments are used the clustering techniques in
order to extract the task-relevant diagnostic group from
psychiatric reports. Typically, the reports comprise psy-
chiatrist’s biographic, referral information, a description
of indications and a proposition on treatment recommen-
dations.
The data set are divided into three categories, first is
the training data set which used to in training of network
to produce the best neuron weight to generalize the net-
work. The second portion is called cross validation data
sets which used to compute the error in a test set at the
same time that the network is being trained with the
training set. Lastly, we tagged some of the data set as
"Testing" to test the performance of the best network. The
weights of best network are automatically saved during
training phase and will be loaded into network before the
testing process is come over.
The experiments and results are performed using Neu-
roSolutions package which is used to design and imple-
ment the MLP network. There are several ways to test the
networks performance [5]. Usually, the mean squared
error MSE is used. It is a function of two times the aver-
age cost and it is computed as follows:
Two hundred messages from each "sci.psychology
psychotherapy" and "sci.philosophy" are selected ran-
domly and made unidentified by deleting header and
signature information [1,10]. The two newsgroups are
chosen according to the previous discusses ideas in psy-
chotherapy including diagnosis and the later may include
similar content but generally must be dissimilar. The clas-
sification mission is difficult since the topic area of the
two newsgroups is so similar that cross-posting can oc-
cur.
The experiments are performed using 200 text files in-
clude the original message plus replies. Further a pre-
processing is implemented on the text samples in order to
encoding them into a binary form which can be used by
the network. Therefore, Inputs are transcribed speech
samples in binary from. And, the output will determine
whether the text has been written by someone who has
mental problems by searching for key symptoms of men-
tal illness in the text such as attention, deficit, hyperactivi-
ty, and disorder (ADHD) [1,10,11].
The experiments are used the clustering techniques in
order to extract the task-relevant diagnostic group from
psychiatric reports. Typically the reports comprise psy-
chiatrist’s biographic, referral information, a description
of indications and a proposition on treatment recommen-
dations.
The data set are divided into three categories, first is
the training data set which used to in training of network
to produce the best neuron weight to generalize the net-
work. The second portion is called cross validation data
sets which used to compute the error in a test set at the
same time that the network is being trained with the
training set. Lastly, we tagged some of the data set as
"Testing" to test the performance of the best network. The
weights of best network are automatically saved during
training phase. They will be loaded into network before
the testing process is come over.

The experiments and results are performed using Neu-
roSolutions package which is used to design and imple-
ment the MLP network. There are several ways to test the
networks performance [5]. Usually, the mean squared
error MSE is used. It is a function of two times the aver-
age cost and it is computed as follows:




Where, P is the number of output processing elements.
N is the number of exemplars in the data set. yij is the
network output for exemplar i at processing element j . dij
is the desired output for exemplar i at processing ele-
TABLE 1
MINIMUM AND FINAL MSE
Best Networks Training Cross Validation
Run NO. 2 2
Epoch NO. 1000 1000
Minimum MSE 0.006527112 0.004332986
Final MSE 0.006527112 0.004332986
( )
) 4 ....(
0
2
NP
y d
MSE
P
o j
N
i
ij ij ¿ ¿
= =
÷
=
JOURNAL OF COMPUTING, VOLUME 4, ISSUE 1, JANUARY 2012, ISSN 2151-9617
https://sites.google.com/site/journalofcomputing
WWW.JOURNALOFCOMPUTING.ORG 8

ment.

The outputs of the network for three runs are summa-
rized in table 1, which records the minimum and final
MSE. The figures 2 and 3 depict the graphs of best net-
work of training and cross validation data respectively.
The average of final MSE standard deviation for train-
ing data set is 0.00186778, while the average of final MSEs
standard deviation for cross validation data set is
0.004812919. With the aim of testing the effectiveness of
the proposed network, the experiments are conducted
using 130 data sets in the testing phase which they ex-
tracted from the total data.
The results of testing the network are epitomized in
Table 2. It shows clearly that all the categories are identi-
fied and classified well. Consequently, the proposed
neural network achieves high accuracy of (99.24242424) in
classification task and effectiveness as measured by the
concepts of precision and recall of (99, 97).















TABLE 2
THE RESULT OF TESTING THE NETWORK
Performance O1(ADHD) O2(attention) O3(deficit) O4(hyperactivity) O5(disorder) O6(None)
MSE 0.008864129 0.001620307 0.007450246 0.003469238 0.002434407 0.010621616
NMSE 1.241042302 0.114255302 1.043088398 0.485718423 0.171661218 0.222099832
MAE 0.050419407 0.033209573 0.0248229 0.051404984 0.047304042 0.057907021
Min Abs Error 0.001047404 0.000876838 0.000873395 0.001632033 0.00215584 0.002345601
Max Abs Error 0.699212803 0.188247272 0.987224796 0.240824599 0.057911202 0.99331302
Correlation 0.672351258 0.968906641 0.695660346 0.871935884 0.982851175 0.911092846
Percent Correct 100 100 100 100 100 99.24242424

TABLE 3
COMPARISON RESULTS OF PROPOSED APPROACH WITH OTHER RESEARCHERS

Kiritchenko [22] Kiritchenko [22]
Joachim [10]
Mabruk [1] Our MLP
Method Naive Bayes SVM SVM SVM NN
Data size for
training phase
100% 100% 100% 100% 20%
Accuracy 80.36% 90.06% 80% 92% 99%
JOURNAL OF COMPUTING, VOLUME 4, ISSUE 1, JANUARY 2012, ISSN 2151-9617
https://sites.google.com/site/journalofcomputing
WWW.JOURNALOFCOMPUTING.ORG 9

7 COMPARISON AND CONCLUSION
7.1 Comparison with related work
The results of the comparison of our work with other re-
searcher related mental health [1,10,11,22] are summarized
in table 3. The current paper using the same data text that
used in the [1,10], this in order to test the experiments under
the same environment and conditions. Our MLP achieved
accuracy of 99%, while Mabruk [1] SVM achieved accuracy
of 92%. And Kiritchenko [22] SVM achieved accuracy of
90.06%. Our approach used a small amount of data to per-
form the training phase, while other approaches use full data
sets to perform the training phase. Consequently, our ap-
proach achieved a highest accuracy, use small amount of
data for training the network and it is a fast approach to
achieve the classification task.

7.2 Conclusion
The increase in mental health research has lead to rise of
information in sake of deterring the problem and found
solutions. However, the precise reasons of a lot of mental
illnesses still unclear. It has been demonstrated that men-
tal illness is a fundamental aspect in many chronic condi-
tions such as diabetes, hypertension, HIV/AIDS resulting
in higher cost to the health system [9,10]
This study is part of a sequence of experiments that
aims to detecting mental illness problems by analyzing
speech, and text. We design and demonstrate the MLP
network to analyze transcribed speech samples of
patients not know if any of the newsgroup messages were
written by people with a mental illness problem.
Nevertheless, the robustness of the machine learning
classification is important which is implemented in this
paper. It has been shown that the MLP approach is
reliable and achieved high accuracy (99%) in classification
of the input text, if it is a type of mental illness or not.
Besides, it used a small amount of data to train the
network which is useful in experiments with rare data.
8 FUTURE WORK
The experiments are proved that more attention must
consecrate to implement the preprocessing task of data
automatically, which includes the reading, encoding, and
clustering. In addition, the researchers can implement
other soft computing techniques such as hybrid of ANN
and genetic algorithms in sake of enhancing the results.
And unsupervised learning approach like Self-organizing
feature maps (SOFMs) to perform the automatic text clas-
sification can be used.
REFERENCES
[1] M. Abusaa, J. Diederich, and A. Al Ajmi, (2004). Web mining and men-
tal health. In: , IAWTIC 2004 Proceedings. International Conference on
Intelligent Agents, Web Technologies and Internet Commerce, Gold
Coast, Queensland,. 12-14 July 2004. http://joachimdiederich.com/-
assets/IAWTIC04.pdf
[2] H. Adelman and Taylor, L. (2006). The current status of mental health
in schools: A policy and practice analysis. Los Angeles, CA: Center for
Mental Health in Schools at UCLA. http://smhp.psych.ucla.edu/-
currentstatusmh.htm
[3] D. E. Akyol (2004). Applications of neural networks to heuristic sche-
duling algorithms. Computers and Industrial Engineering, 46, 679-696.
http://www.sciencedirect.com/science/article/pii/S036083520400066
X
[4] A. Ben-Hur, D. Horn, H. T. Siegelmann and Vapnik, V. (2001). Support
vector clustering. Journal of Machine Learning Research, 2, 125–137.
http://jmlr.csail.mit.edu/papers/volume2/horn01a/rev1/horn01ar1.
pdf
[5] H. G. Beyer and H. P. Schwefel (2002). Evolution strategies: A compre-
hensive introduction. Natural Computing: An International Journal,
1(1), 3–52. DOI: 10.1023/A:1015059928466
[6] P. Bonissone (2002). Hybrid Soft Computing for Classification and
Prediction Applications. Conferencia Invitada. 1st International Confe-
rence on Computing in an Imperfect World (Soft-Ware 2002), Belfast.
DOI: 10.1007/3-540-46019-5_28
[7] G. Clarke, F. Lynch , M. Spofford and L. DeBar (2006). Trends influen-
cing future delivery of mental health services in large healthcare sys-
tems. Clinical Psychology: Science and Practice, 13 (3) , 287-292.
doi:10.1111/j.1468-2850.2006.00040.
[8] M. Hadzic, M. Chen and T. Dillon (2008). “Towards the mental health
ontology”, Proceeding of the IEEE International Conf. on Bioinformat-
ics and Biomedicine, USA. DOI: 10.1109/BIBM.2008.59
[9] E. J. S. Hovenga (2004). “Globalisation of health and medical informat-
ics education: What are the issues?” International Journal of Medical In-
formatics, 73(2), 101–109. DOI: 10.1016/j.ijmedinf.2003.11.004
[10] J. Diederich , et.al, (2007). Ex-ray: “Data mining and mental health”,
Applied Soft Computing 7 (2007) 923–928. doi:10.1016/j.asoc.-
2006.04.007.
[11] J. Diederich , P. Yellowlees, A. Al-Ajmi. Ex-Ray: “Text Classification and
the Assessment of Mental Health”, P. Bruza, J. Thom. (Eds.), Proceed-
ings of the Eighth Australian Document Computing Symposium, Can-
berra, Australia (15 December 2003) 75–76. Canberra: CSIRO Discovery
Centre (2003). http://joachimdiederich.com/assets/ExRay-
ADCS03.pdf
[12] R. Lippman “An introduction to computing with neural nets”. IEEE
Trans. ASSP Magazine 4, 4-22, 1987.
[13] H. Lodhi, C. Saunders, J. Shawe-Taylor, N. Cristianini, N. and C.
Watkins, C. (2002). Text classification using string kernels. Journal of
Machine Learning Research, 2, 419–444. Paris: EC2 & Cie.
http://citeseer.ist.psu.edu/viewdoc/summary?doi=10.1.1.130.2853
[14] J. Makhoul “Pattern recognition properties of neural networks”. Proc.
1991 IEEE Workshop on Neural Networks for Signal Processing, pp
173-187, 1992.
DOI: 10.1109/NNSP.1991.239524
[15] H. Nazeran and K. Behbehani (2001). Neural networks in processing
and analysis of biomedical signals. In M. Akay (Ed.), Nonlinear bio-
medical signal processing: Fuzzy logic, neural networks and new algo-
rithms (pp. 69–97).
[16] V. Patel, M. Prince (2010). “Global mental health - a new global health
field comes of age”. JAMA, 303, 1976-1977. doi: 10.1001/jama.2010.616
[17] A. Power (2010), “Transforming the Nation's Health: Next Steps in
Mental Health promotion”, American Journal of Public Health 100 (12):
2343.
DOI: 10.2105/AJPH.2010.192138
[18] K. C. Richards, C. Campania, J. L .Muse-Burke (2010). “Self-care and
Well-being in Mental Health Professionals: The Mediating Effects of
Self-awareness and Mindfulnes”. Journal of Mental Health Counsel-
JOURNAL OF COMPUTING, VOLUME 4, ISSUE 1, JANUARY 2012, ISSN 2151-9617
https://sites.google.com/site/journalofcomputing
WWW.JOURNALOFCOMPUTING.ORG 10

ing 32 (3): 247.
http://www.thefreelibrary.com/Self-care +and+well-
being+in+mental+health+ professionals%3A+the...-a0233291361
[19] R. Rifkin and A. Klautau (2004). In defense of one-vs.-all classification.
Journal of Machine Learning Research, 5, 101–141.
http://cbcl.mit.edu/projects/cbcl/publications/ps/rifkin-04-jmlr.pdf
[20] Rumelhart D. E., G. E. Hinton, and R. J. Williams. Learning internal
representations by error propagation. In D. E. Rumelhart and J. L.
McClelland, editors, Parallel Distributed Processing, volume 1, pages
318–362, Cambridge, USA, 1986. MIT-Press.
[21] Sejnowski T. and C. Rosenberg. Parallel networks that learn to pro-
nounce English text. Complex Systems, pages 145–168, 1987.
[22] S. Kiritchenko and S. Matwin. Email Classification with Co-Training, in
Proceedings of the 2001 Conference of the Centre for Advanced Studies
on Collaborative Research (CASCON). Toronto, Canada, 2001.
[23] Smith, K & Gupta J., Neural networks in business: Techniques and
applications for the operations researcher, Computer & Operation Re-
search, (27), pages 1023-1044, 2000.
[24] Zadeh, L.A. (2001). Applied Soft Computing. Applied Soft Computing
1, 1–2.
JOURNAL OF COMPUTING, VOLUME 4, ISSUE 1, JANUARY 2012, ISSN 2151-9617
https://sites.google.com/site/journalofcomputing
WWW.JOURNALOFCOMPUTING.ORG 11

Sponsor Documents

Or use your account on DocShare.tips

Hide

Forgot your password?

Or register your new account on DocShare.tips

Hide

Lost your password? Please enter your email address. You will receive a link to create a new password.

Back to log-in

Close