Dynamic Rule Base Construction and Maintenance Scheme for Disease Prediction

IJSRD - International Journal for Scientific Research & Development| Vol. 1, Issue 5, 2013 | ISSN (online): 2321-0613
All rights reserved by www.ijsrd.com 1246
Abstract— Business and healthcare application are tuned to
automatically detect and react events generated from local
are remote sources. Event detection refers to an action taken
to an activity. The association rule mining techniques are
used to detect activities from data sets. Events are divided
into 2 types’ external event and internal event. External
events are generated under the remote machines and deliver
data across distributed systems. Internal events are delivered
and derived by the system itself. The gap between the actual
event and event notification should be minimized. Event
derivation should also scale for a large number of complex
rules. Attacks and its severity are identified from event
derivation systems. Transactional databases and external
data sources are used in the event detection process.
The new event discovery process is designed to support
uncertain data environment. Uncertain derivation of events
is performed on uncertain data values. Relevance estimation
is a more challenging task under uncertain event analysis.
Selectability and sampling mechanism are used to improve
the derivation accuracy. Selectability filters events that are
irrelevant to derivation by some rules. Selectability
algorithm is applied to extract new event derivation. A
Bayesian network representation is used to derive new
events given the arrival of an uncertain event and to
compute its probability. A sampling algorithm is used for
efficient approximation of new event derivation.
Medical decision support system is designed with event
detection model. The system adopts the new rule mapping
mechanism for the disease analysis. The rule base
construction and maintenance operations are handled by the
system. Rule probability estimation is carried out using the
Apriori algorithm. The rule derivation process is optimized
for domain specific model.
I. INTRODUCTION
Medical data mining has great potential for exploring the
hidden patterns in the data sets of the medical domain.
These patterns can be utilized for clinical diagnosis.
However, the available raw medical data are widely
distributed, heterogeneous in nature, and voluminous. These
data need to be collected in an organized form. This
collected data can be then integrated to form a hospital
information system. Data mining technology provides a user
oriented approach to novel and hidden patterns in the data.
Healthcare has been one of the top demands of this
generation. In the advent of technology, the provision of
healthcare continues to improve. One of the top priorities is
the provision of diagnosis, and the one responsible for this is
the physician. Physician’s diagnosis is the most relevant
factor that leads to the acquisition of proper health guides.
As technology soars, there are changing medical
requirements and solutions, and sometimes physicians are
not updated with these upgrades, that they need to surf the
internet for more information.
Medical diagnosis is regarded as an important yet
complicated task that needs to be executed accurately and
efficiently. The automation of this system would be
extremely advantageous. Regrettably all doctors do not
possess expertise in every sub specialty and moreover there
is a shortage of resource persons at certain places.
Therefore, an automatic medical diagnosis system would
probably be exceedingly beneficial by bringing all of them
together. Appropriate computer-based information and/or
decision support systems can aid in achieving clinical tests
at a reduced cost. Efficient and accurate implementation of
automated system needs a comparative study of various
techniques available.
One of the solutions that could aid in the physician’s
diagnosis is the Clinical Diagnostics Decision Support
System (CDSS). The CDSS is generally defined as any
computer program designed to help health professionals
make clinical decisions. Clinical Decision Support Systems
supports case-specific advice which addresses to the aiding
of physician’s diagnosis via computer-based facility. Since
physicians sometimes encounter complicated ailments, a
CDSS can make decision making simpler by providing
relevant pre-diagnosis. With the use of CDSS, a pre-
diagnosis could be extracted which can strongly support the
physician’s decision. Hence, lesser time could only be
consumed in the provision of diagnosis to the patient. Along
with this CDSS is a logic which provides probabilistic
conditions that leads to a specific outcome, which is the
Fuzzy Logic; another one is the method to be taken, which
is the rule-based method.
II. RELATED WORK
Complex event processing is supported by systems from
various domains. These include ODE, Snoop and others for
active databases and the Situation Manager Rule Language,
a general purpose event language. Event management was
also introduced in the area of business process management
[3] and service-based systems [7]. An excellent introductory
book to complex event processing is also available. A recent
book introduces principles and applications of distributed
event based systems [6]. Architectures for complex event
processing were proposed, both generic and by extending
middleware.
The majority of existing models do not support event
uncertainty. Therefore, solutions adopted in the active
database literature, such as the Rete network algorithms fail
to provide an adequate solution to the problem, since they
cannot estimate probabilities. An initial rule-based approach
for managing uncertain events was proposed in [14]. This
work was followed by a probabilistic event language for
Dynamic Rule Base Construction and Maintenance Scheme for
Disease Prediction
M. Mythili1
Prof. K. Sampath kumar2
1
ME (CSE), 2
HOD/CSE
PGP College of Engineering and Technology, Namakkal

Dynamic Rule Base Construction and Maintenance Scheme for Disease Prediction
(IJSRD/Vol. 1/Issue 5/2013/0051)
supporting RFID readings. Our proposed model shares some
commonalities with, namely uncertainty specification at
both the event occurrence and the rule level, and providing
an algorithm for uncertain event derivation. However, unlike
[4], our framework does not limit the type of temporal
expressions it can handle nor does it limit attribute
derivation types. We extend the work by providing
algorithms for deriving uncertain events and empirical
evidence to the scalability of the approach.
Recent works on event stream modelling propose formal
languages to support situations such as event negations.
Various aspects of event-oriented computing were discussed
at CIDR 2007 (e.g., [2]). Our focus in this work is on
modelling and efficiently managing uncertainty in complex
event systems, which was not handled by these works.
Modelling probabilistic data and events was suggested in
[8]. This work extends those models by proposing a model
and a probability space representation of rule uncertainty. A
common mechanism for handling uncertainty reasoning is a
Bayesian network, a method for graphically representing a
probability space, using probabilistic independencies to
enable a relatively sparse representation of the probability
space. Qualitative knowledge of variable interrelationships
is represented graphically, while quantitative knowledge of
specific probabilities is represented as Conditional
Probability Tables (CPTs). The network is, in most
applications, manually constructed for the problem at hand.
In our framework, we automatically construct a Bayesian
network from a set of events and rules, following the
Knowledge Based Model Construction (KBMC) paradigm.
This paradigm separates uncertain knowledge representation
from inference, which is usually carried out by transforming
the knowledge into a Bayesian network that can model
knowledge at the propositional logic level knowledge. The
work also follows the KBMC paradigm. In this work, the
quantitative knowledge, as well as the quantitative
deterministic knowledge, are represented as a set of Horn
clauses and the qualitative probabilistic knowledge is
captured as a set of CPTs. Our framework, however, in that
our framework need not be restricted to first order
knowledge. For example, second order knowledge, or
knowledge that can be expressed using any procedural
language, can also be captured. In addition, in our
framework, probabilities may have a functional dependency
on the events themselves. For example, the probability
assigned to a flu outbreak could be specified to be min (90 +
(X - 350)=10, 100). In addition, our framework enables the
uncertainty associated with such event derivation to be
captured by different probability spaces at different points in
time. Existing works [9] for processing complex events in
specific domains tailor probabilistic models or direct
statistical models to the application. No general framework
was defined there to derive uncertain events. Some
details of our framework were presented in [13] with two
significant extensions in this work. We discuss in details
efficiency aspects of our algorithms, and in particular event
Selectability. We also significantly extended our empirical
analysis to include more cases.
III. EVENT DRIVEN SYSTEMS (EDS)
In recent years, there has been a growing need for event
Driven systems, i.e., systems that react automatically to
events. The earliest event-driven systems in the database
realm impacted both industry and academia. New
applications in areas such as Business Process Management
(BPM) [5]; sensor networks [11]; security applications;
engineering applications; and scientific applications all
require sophisticated mechanisms to manage and react to
events.
Some events are generated externally and deliver data across
distributed systems, while other events and their related data
need to be derived by the system itself, based on other
events and some derivation mechanism. In many cases, such
derivation is carried out based on a set of rules. Carrying out
such event derivation is hampered by the gap between the
actual occurrences of events, to which the system must
respond, and the ability of event-driven systems to
accurately generate events. This gap results in uncertainty
and may be attributed to unreliable event sources, an
unreliable network, or the inability to determine with
certainty whether a phenomenon has actually occurred given
the available information sources. Therefore, a clear trade-
off exists between deriving events with certainty, using full
and complete information, and the need to provide a quick
notification of newly revealed events. Both responding to a
threat without sufficient evidence and waiting too long to
respond may have undesirable consequences.
In this work, we present a generic framework for
representing events and rules with uncertainty. We present a
mechanism to construct the probability space that captures
the semantics and defines the probabilities of possible
worlds using an abstraction based on a Bayesian network. In
order to improve derivation efficiency we employ two
mechanisms: The first mechanism, which we term
Selectability, limits the scope of impact of events to only
those rules to which they are relevant, and enables a more
efficient calculation of the exact probability space. The
second mechanism we employ is one of approximating the
probability space by employing a sampling technique over a
set of rules. We show that the approximations this
mechanism provides truly represent the probabilities defined
by the original probability space. We validate the
approximation solution using a simulation environment,
simulating external event generation, and derive events
using our proposed sampling algorithm.
IV. CHALLENGES IN EDS
A generic model is used for representing the derivation of
new events under uncertainty. Uncertain derivations of
events are performed on uncertain data values. Relevance
estimation is a more challenging task under uncertain event
analysis. Selectability and sampling mechanism are used to
improve the derivation accuracy. Selectability filters events
that are irrelevant to derivation by some rules. Selectability
algorithm is applied to extract new event derivation. A
Bayesian network representation is used to derive new
events given the arrival of an uncertain event and to
compute its probability. A sampling algorithm is used for
efficient approximation of new event derivation. The
following drawbacks are identified from the existing system.
1. Static rule base model
2. Limited accuracy in rule probability estimation

3. Manages limited incoming events only
4. Rule class and structure are not generalized
V. EVENT DERIVATION MODELS
The challenges associated with event derivation; note first
that such events only suggest a high probability of a flu
outbreak, which does not necessarily mean that such an
event should be derived. We term this type of uncertainty
derivation uncertainty, as it stems from the inability to
derive such events with certainty from the available
information. In addition, there is uncertainty regarding the
data itself, because the provided data are rounded to the
nearest ten. For example, the increase from November 30 to
December 4 is of 350 units, yet rounding may also suggest a
smaller increase of 341 units. This is considered uncertainty
at the source, resulting from inaccurate information
provided by the event source. Additional details regarding
the different types of uncertainty can be found in [1].
EID Date Daily Sales (Rounded)
113001 Nov 30 600
120101 Dec 1 700
120201 Dec 2 800
120301 Dec 3 900
120401 Dec 4 950
120501 Dec 5 930
Table 1: Over-The-Counter Sales Relation
Fig. 1: Anthrax rule
To highlight the complexity of the processing uncertain
events and rules we note that the rule presented above is a
simplified version of rules expected to exist in the real
world. For example, instead of just specifying a single
certainty level of 90 percent, one could compute the
probability of an outbreak as a function of the amount of
increase. Note that while such a rule may be sufficient to
capture in some settings the derivation uncertainty, it does
not capture uncertainty at the source.
We also note that real-life applications need to
process multiple rules with multiple data sources [10]. To
illustrate, consider Fig. 1, where we need to consider the
possibility of an anthrax attack, in addition to the flu
outbreak. In this setting, data must be combined from events
that come from different data sources. A lower probability
should be assigned to an anthrax attack whenever a flu
outbreak is recognized and a higher probability otherwise.
As a result, a flu outbreak event is not derived; a significant
increase of hospital emergency department visits with
respiratory complaints is enough to assign a probability of
60 percent to the occurrence of an anthrax attack event.
Other examples of more complex settings may involve
correlating data between pharmacies or across regions.
Therefore, it must be possible to specify a set of rules that
captures this complexity.
VI. PROBABILISTIC EVENT MODEL
A. Event Model
An event is an actual occurrence or happening that is
significant and atomic. Examples of events include
termination of workflow activity, daily OtCCMS and a
person entering a certain geographical area. We differentiate
between two types of events. Explicit events are signaled by
external event sources. Derived events are events for which
no direct signal exists, but rather need to be derived based
on other events, e.g., Flu Outbreak and Anthrax Attack
events. Data can be associated with an event occurrence.
Some data types are common to all events, while others are
specific. The data items associated with an event are termed
attributes.
B. Derivation Model
Derived events in our model are inferred using rules. For
ease of exposition, we refrain from presenting complete rule
language syntax. Rather, we represent a rule by a quintuple,
r = defining the necessary conditions for the derivation of
new events. Such a quintuple can be implemented in a
variety of ways, such as a set of XQuery statements, Horn
clauses, and CPTs such as in [12], or as a set of procedural
statements. We detail next each of the rule elements,
illustrating them with the rule r1: “If there is an increase in
OtCCMS for four sequential days to a total increase of 350,
then the probability of a flu outbreak is 90 percent.” Recall
that OtCCMS events contain the volume of the daily sales.
C. Event Detection Algorithms
Selectability, as defined by function sr in a rule
specification, plays an important role in event derivation, in
both the deterministic and the uncertain settings. In our
setting, the relevance of an EID to derivation according to
rule r is determined by the function sr. sr is a function sr : h
hr that receives an event history as input, and returns a
subset hr h. As an example of a function sr consider the
selection function corresponding. Given an event history h,
it returns a set consisting of all possible events e that
indicate an increase in OtCCMS in two consecutive days.
Formally, this would be defined as all possible events e such

that e may be one of a pair of events {e1, e2} in h such that
both e1 and e2 are OtCCMS events, e2 occurred one day
after e1, and the number of cough medication sold as
indicated by e2 is greater than the number of cough
medication sold as indicated by e1. It is the role of the
selection expression to filter out events that are irrelevant for
derivation according to r. An event e h to be relevant to
derivation according to rule r iff e sr (h).We also require
that for every pair of event histories h, h’ such that h h0 sr
(h) it must hold that sr (h) = sr (h’). (1)
Note that from (1) we have the following special case: sr
(h) = sr (sr (h)). (2)
1) Selectability Algorithm
As in the uncertain setting derivation is carried out on EIDs,
algorithms are required to compute which EIDs, from a
given system event history H, are selectable. Deciding
whether an EID E is selectable by rule r may, by itself, incur
significant computational effort. This is because,
selectability depends on the possible event histories in
which the event corresponding to E participates. Therefore,
a naive algorithm for finding all selectable EIDs involves
scanning all event histories, and for each event history,
finding all events selectable by rule r using the function sr.
However, this may be extremely inefficient, as the number
of possible event histories may be exponential in the size of
the state space of the events. Therefore, if the system event
history H contains n EIDs, and the size of the state space of
the largest EID is m, then the number of possible event
histories is O(mn).
We propose an efficient algorithm based on a decomposition
of sr. We decompose sr into two functions as follows: Let =
{e1 ,. . . , en}, then sr may be described by csr (esr (e1)) esr
(e2) ….. esr (en)), such that esr : e {{}, {e}.esr receives a
single event as input, and returns as output either the empty
set if it is clear from the attributes of this event that this
event is not selectable by the rule, or the set containing the
single event received as input otherwise. This representation
enables distinguishing between a selection that can be
carried out only by looking at an individual event and its
attributes using the function e sr, and filtering that requires
looking at combinations of events in h, which is carried out
by function c sr. Note that such decomposition into e sr and
c sr can always be carried out, since e sr can, in the worst
case, return the set containing the input event. In such a
case, sr = csr.
Based on this decomposition we provide an algorithm that,
for each rule r, provides the subset of EIDs in a system event
history that is selectable by r. In this algorithm the function
esr is first used to test selectability at the individual EID
level. A set of EIDs will be constructed such that EID E is in
the set iff one of the possible states of E corresponds to an
event e that esr returns as selectable. Following this, only
the set of event histories defined by this subset of EIDs is
checked for selectability, by using the c sr function. The
pseudocode of this algorithm—function
calculateSelectableEIDs—appears in Algorithm 1.
Algorithm 1. calculateSelectableEIDs(H,r)
1. selectableEIDs
2. esEIDs ,
3. for all E H
4. for all e E
5. if esr (e) = {{e}}
6. esEIDs esEIDs E
7. end if
8. end for
9. end for
10.for all h esEIDs
11. for all e sr (h)
12. E getCorrespondingEID(e)
13. selectableEIDs selectableEIDs E
14. end for
15. end for
16. return selectableEIDs
2) Sampling Algorithm
The algorithm described in the generates a Bayesian
network from which the exact probability of each event can
be computed. Given an existing Bayesian network, it is also
efficiently possible to approximate the probability of an
event occurrence using a sampling algorithm, as follows:
Given a Bayesian network with nodes E1, . . . , En, we
calculate an approximation for the probability that Ei =
{occurred} by first generating m independent samples using
a Bayesian network sampling algorithm. Then, Pr(Ei =
{occurred}) is approximated by , where #(Ei = {occurred})
is the number of samples in which Ei has received the value
occurred.
Algorithm 2. RuleSamp, triggered by a new event arrival
1. h
2. for all E H0
3. e probSampling(E)
4. h h e
5. end for
6. order obtainTriggeringOrder
7. while order
8. r deleteNextRule(order)
9. h’
10. selEvents sr (h)
11. if pr(selEvents) = true
12. assocT uples ar (selEvents)
13. for all tuple assocT uples
14. {s1, . . . , sn} mr (tuple)
15. prob prr (tuple, mr (tuple))
16. probSamp
sampleBernoulli(prob)
17. if probSamp = 1
18. e {occurred, s1,. , sn}
19. else
20. e {notOccurred}
21. end if
22. h h’ {e}
23. end for
24. end if
25. h h h’
26. end while
27. return h
VII. DISEASE PREDICTION WITH HYBRID RULE BASE
MODEL
The event derivation system is enhanced to map dynamic
rules on uncertain data environment. The rule base
construction and maintenance operations are handled by the
system. Rule probability estimation is carried out using the

Apriori algorithm. The rule derivation process is optimized
for domain specific model. The system is designed to detect
events on uncertain data environment. Dynamic rule base
model is used in the system. The system integrates the rule
base update process. The system is divided into six major
modules. They are patient diagnosis, rule base management,
sampling process, selection process, event detection and rule
base update.
The patient diagnosis module is designed to collect patient
symptoms. The rule base is constructed with experts support
under rule base management module. The sampling process
module is designed to select item samples. The selection
process module is designed to select item combinations. The
rule base comparison is performed under event detection
module. The rule base update module is designed to update
new events with rules.
A. Patient Diagnosis
The system uses TB patient diagnosis information. Patient
diagnosis data can be imported from external databases. The
user can also update new patient diagnosis details. Diagnosis
list shows the patient symptom levels.
B. Rule Base Management
Rule base is used to manage the rules and event names. Rule
base details are collected from domain experts. The rules are
composed with attribute combination and values. Three
types of rules are used in the system. Static, dynamic and
hybrid rules are maintained under the rule base.
C. Sampling Process
The sampling process is performed to select event derivation
models. The attribute values are sampled from uncertain
data collections. The sampling algorithm is used to select
sample data derivations. Sampled data values are passed to
rule analysis process.
D. Selection Process
The selection process is applied to filter irrelevant events.
Item combinations are selected under the selection process.
The Selectability algorithm is used in the selection process.
Event derivations are used in the rule analysis process.
E. Event Detection
The rule base analysis is performed with user data
collections. The Bayesian network is used in the event
detection process. The attribute combinations are compared
with rule base information. Event detection is carried out
with rank and priority information.
F. Rule Base Update
The dynamic rules are derived from user data and static rule
base information. The dynamic rules are updated with event
name and priority values. Rule ranking is performed with
frequency information. Rules are updated with priority
details.
VIII. CONCLUSION
Rule bases are built to manage activities and their class
information. Event derivation is carried out with rule bases.
Dynamic rule base update model introduced to improve the
event detection process. Selective and sampling algorithms
are enhanced to derive events under dynamic and uncertain
domain environments. The system integrates the experts
knowledge and local domain information for rule base
construction. Rule base generation and maintenance
operations are done using machine learning models. Event
classes and its structures are generalized. Association rule
mining methods are used to extract rule patterns.
REFERENCES
[1] A. Gal, and O. Etzion, A. Taxonomy,
“Representation of Sources of Uncertainty in Active
Systems,” Proc. Sixth Int’l Conf. Next Generation
Information Technologies and Systems, 2006.
[2] L. Girod, R. Newton, S. Rost, A. Thiagarajan, H.
Balakrishnan, and S. Madden, “The Case for a
Signal-Oriented Data Stream Management System,”
Proc. Conf. Innovative Data Systems Research, 2007.
[3] R. Ammon, C. Emmersberger, T. Greiner, A.
Paschke, F. Springer, and C. Wolff, “Event-Driven
Business Process Management,” Proc. Second Int’l
Conf. Distributed Event-Based Systems, July 2008.
[4] M. Balazinska, N. Khoussainova, and D. Suciu,
“PEEX: Extracting Probabilistic Events from Rfid
Data,” Proc. Int’l Conf. Data Eng. (ICDE), 2008.
[5] C. Beeri, T. Milo, and A. Pilberg, “Query-Based
Monitoring of BPEL Business Processes,” Proc.
ACM SIGMOD Int’l Conf., 2007.
[6] Principles and Applications of Distributed Event-
Based Systems, A. Hinze and A. Buchmann, eds. IGI
Global, 2010.
[7] H. Chen and J. Dong, “Jtang Synergy: A Service
Oriented Architecture for Enterprise Application
Integration,” Proc. 11th Int’l Conf. Computer
Supported Cooperative Work in Design, Apr. 2007.
[8] A. Demers, M. Riedewald, V. Sharma, and W. White,
“Cayuga a General Purpose Event Monitoring
System,” Proc. Third Biennial Conf. Innovative Data
Systems Research, pp. 412-422, 2007.
[9] M. Campbell, C.-S. Li, C. Aggarwal, M. Naphade,
K.-L. Wu, and T. Zhang, “An Evaluation of over-the-
Counter Medication Sales for Syndromic
Surveillance,” Proc. IEEE Int’l Conf. Data Mining—
Life Sciences Data Mining Workshop, 2004.
[10] Segev Wasserkrug, Opher Etzion, and Yulia Turchin,
“Efficient Processing of Uncertain Events in Rule-
Based Systems” IEEE Transactions On Knowledge
and Data Engineering, January 2012.
[11] Hellerstein, and W. Hong, “Approximate Data
Collection in Sensor Networks Using Probabilistic
Models,” Proc. 22nd Int’l Conf. Data Eng , 2006.
[12] K. Kersting and L. De Readt, “Bayesian Logic
Programming,” An Introduction to Statistical
Relational Learning, pp. 291-322, MIT Press, 2007.
[13] S. Wasserkrug, and Y. Turchin, “Complex Event
Processing over Uncertain Data,” Proc. Second Int’l
Conf. Distributed Event-Based Systems, 2008.
[14] S. Wasserkrug, A. Gal, and O. Etzion, “A Model for
Reasoning with Uncertain Rules in Event
Composition Systems,” Proc. 21st Conf. Uncertainty
in Artificial Intelligence, 2005.

Dynamic Rule Base Construction and Maintenance Scheme for Disease Prediction

More Related Content

What's hot (20)

Viewers also liked (18)

Similar to Dynamic Rule Base Construction and Maintenance Scheme for Disease Prediction (20)

More from ijsrd.com (20)

Recently uploaded (20)

Dynamic Rule Base Construction and Maintenance Scheme for Disease Prediction