A PROPOSED DSS FOR DIABETES PREDICTION
3.1 THE OVERVIEW OF PROPOSED DSS
A knowledge based decision
support system (KD-DSS) designed and developed using standard DSS Model that
works to predict diabetes. From input to output DSS evolved an Inference Engine
and a Knowledge Construction mechanism inside. To work perfectly two well-known
algorithms C4.5 and SVM implemented in proposed DSS as a predictor techniques.
DSS trained on Pima Indians Diabetes Dataset and proposed DSS designed and
developed using C# with an easy to use user interface. Following fig. 3.1 is
the system architectural elaboration of proposed DSS.
3.1 System Architectural Diagram
Input for prediction is new record with biological information of any
person to predict the diabetes orientation in future. Input could be 8
parameters according to the dataset on which decision support system trained.
Inference engine then apply the rules and map the rules with real readings so
that knowledge construction could be done in descriptive way. Knowledge
construction trained the model on available historical data items in Pima
Indians diabetes data set. Output is descriptive information generated by
proposed decision support system that produces descriptive output.
3.2 CONCEPTUAL DIAGRAM OF PROPOSED DSS
Proposed DSS implemented in different steps that are
portioned logical in different phases. These phases from data codification to
splitting the attributes till final prediction are linked and work
collaboratively to produce descriptive output. Fig. 3.2 represent conceptual
diagram of Proposed DSS.
3.2 Conceptual Diagram of Proposed DSS
Since the labels are represented as text, the first step is
to convert those text labels into integer class labels, so we can process them.
This conversion is done in first step that is Data Codification. Next step is
to split the attributes from class to other data attributes. After the
codification and splitting the records decision support system is ready to
apply classifier. We may choose C4.5 or SVM to apply and predict our new
instance. Next step is to decide the prediction that is new instance prediction
based on trained model. System map and evaluate the new instance according to
the generated rules and predict the final output. Predicted output is further
pass for formulation the output. Output formulation use conditions and
formulate the output in such descriptive way that is easily understandable for
3.3 DETAILED DIAGRAM OF PROPOSED DSS
DSS is developed by implementing two well-known algorithms of machine learning
for prediction and train the model. C4.5 proposed by Ross Quinlan in 1993 is
the extension of ID3 decision tree classifier. C4.5 got success and attraction
after ranking #1 of top 10 algorithms in data mining published by Springer in
2008. Sequential Minimal Optimization (SMO) a fast algorithm to train the
Support Vector Machines developed by John Platt in 1998 in Microsoft Research
3.3 Detailed Diagram of Proposed DSS
3.3 presents detailed elaboration of proposed DSS.
System works in steps toward final decision. User has to run the DSS and select
the classifier that is predictor option. We may set C4.5 or SVM as predictor as
start our model to train. Meanwhile in training, system connects to dataset and
start knowledge discovery and produce rules for decision making. After model
successfully trained user may provide new instance for prediction. New instance
is evaluated by generated rules and system invokes the next step to make
decision. Decided output is a class label that needs to formulate in
descriptive manner to understand. Output formulation is the last step that
produces a descriptive output to user.
3.4 PROPOSED DSS DESIGN
Fig. 3.4 is the UI
design of developed DSS in C#. Interface is designed in four parts included
learning that is first phase where user set the algorithm and starts learning.
System also provides dataset characteristics in this phase after learning from
data provided as dataset. To predict diabetes of new patient, system has text
fields available in next to provide as input. New patient readings are taken
from input fields and provided to DSS for decision making. Output is formulated
in easy descriptive manner and available in Output section at interface.
Summary about the prediction and DSS performance is available at the end bottom
Fig. 3.4 Proposed DSS Interface