87 PPT

Measuring Defect Potentials and
Minimizing the Difficulty of SQA

by Automated Techniques
By
H.K. Gundu Rao,
HOD, Department of Computer
science, Vijaya College, R.V Road
Bangalore.
About the work

Machine learning technique for defect forecasting and handling for SQA called appendage
log training and analysis, can be referred as ALTA.
The proposed defect forecasting of in-appendage software development logs is to deal the
forecasted defects accurately and spontaneously while developing the software.
The present proposed mechanism helps in minimizing the difficulty of SQA.
The overall study is conducted on evaluating the proposed model which indicates the defect
forecasting and reduces the complexity of Software Quality Assessment.
Introduction
Numbers of varieties of models are introduced
Reliability growth models are predictive models
Metric based models such as maintainability index is an

assessment model
ISO 9126 model which is definition model
Prediction models are most advanced form of models
Related work
Quality models must be central treasure for the information about quality. So the
various tasks of quality engineering should depend upon one quality model.
Now days, first problem is that quality models are used separately for
specification of quality needs and the assessment of software quality.
Second problem is that now a days quality models are not addressing various
opinions on quality.
For software engineering, the value-based opinion is commonly considered of

high importance which is mostly missing in existing quality models
Software systems consists a large variety which ranges from a large amount of
business information systems to small embedded controllers.
These variations must be accounted for in quality models by defined means of

customization but is lacking in the present quality models
It should be stated that measurement is important for any control

process.
Hence the measurement of quality attributes is necessary for an

efficient quality assurance processes and for engineering needs.
Predictive quality models dont possess definition of the concepts

they are depend on.
Many of them depend on regression using a set of software

metrics. This regression results in equations that are difficult to
interpret
Developers, hence, require new techniques to aid in exploring
the indication of serialized-phasing development and enable
evaluation of QoS that is considered throughout the software .
SQA by defect prediction

The mentioned In-Appendage Software Development log Analysis by SVM works as
an intelligent system to identify defects to enhance the Software Quality. In nut shell
we refer ALTA.
The proposed defect prediction process that mingles with software development
and testing. In ALTA the defect prediction is strategic and helps to predict the defect
that leverages the cost and targeted result.
DEFECT PREDICTION USING LS-SVM AND IWP-QPSO LS-SVM

For solving pattern recognition and classification problem using Support vector
machine (SVM) tool proposed by Vapnik is beneficent.
It even can be applied to regression problems by the brining up other loss
function.
Linear and nonlinear regression can be performed with LSSVM. Therefore finalizing
the SVM model is computationally very hard because it needs the solution of a set
of nonlinear equations .
According to simplification, Suykens and Vandewalle proposed a modified version of

SVM called least-squares SVM (LS- SVM), that bring out in a set of linear equations
in the place of a quadratic programming problem.
Basically, LS-SVM uses a linear relation (y = wx + b) between the regression (x) and
the dependent variable (y).
The performances of LS-SVM based on the combination of several parameters.
IWP- QPSO (Intensifying the Worst Particle for QPSO)

S.Nagaraja Rao did to optimize the QPSO by intensifying worst swarm particle with new
swarm particle. An interpolate equation will be traced out by applying a quadratic
polynomial model on existing best fit swarm particles.
The computational steps of optimized QPSO algorithm as follows:
Step 1: Initialize the swarm.
Step 2: Calculate mbest
Step 3: Update particles position
Step 4: Evaluate the fitness value of each particle
Step 5: If the current fitness value is better than the best fitness value (Pbest) in history
Then Update Pbest by the current fitness value.
Step 6: Update Pgbest (global best)
Step 7: Find a new particle
Step 8: If the new particle is better than the worst particle, then replace the worst
particle by the new particle.
Step 9: Go to step 2 until maximum iterations reached.
The swarm particle can be found using the following.
ti
= ( (p
- qi ) * f(r)
k =1
p = a, q = b, r = c
for k = 1;
p = b, q = c, r = a for k = 2
p = c, q = a, r = b for k = 3
t1i = (pi - qi ) * f (r)

k =1
p = a, q = b, r = c
for k = 1;
p = b, q = c, r = a for k = 2
p = c, q = a, r = b for k = 3
= 0.5 (
)
1
LS-SVM Regression and QPSO based hyper parameter

selection
Assume that a given training set of N data points {xt, yt}nt=1 with input data x
y R . In feature space LS-SVM [25] regression model take the form

y (x) = w j (x) + b (3)
Where the input data is mapped j (.)
and output
In present work, the radial basis function (RBF) is used as the kernel function:
k ( xi , x j ) = exp(- || x - xt ||2 /s 2 )
In the training LS-SVM problem, there are hyper parameters, such as kernel width
parameter and regularization parameter C, which may affect LS-SVM
generalization performance. So these parameters required perfectly in tuning of
reducing the generalization error. An attempt is made to tune these parameters
automatically by using QPSO.
Hyper-Parameters Selection Based on IWP-QPSO

To surpass the usual loss results in least-square SVM, we attempt to optimize
hyper parameter selection.
The two key factors that finalize the optimized hyper-parameters through QPSO:
First one is how to represent the hyper parameters as the particle's position,
namely encoding . Second one is defining the fitness function, that evaluates the
goodness of a particle
Encoding Hyper-parameters:The optimized hyper-parameters for LS-SVM
consists kernel parameter and regularization parameter.
To solve hyper parameters selection by the proposed IWP-QPSO , every particle

represents a potential solution, called hyper-parameters combination. A hyperparameters combination of dimension m is represented in a vector of dimension m
The following steps describe the IWP-QPSO -Trained LSSVM algorithm:

(1) Initialize the population by randomly generating the position vector i and X of each
particle and set iP = iX;
(2) Structure LS-SVM by treating the position vector of each particle as a group of
hyper-parameters;
(3) Train LS-SVM on the training set;
(4) Evaluate the fitness value of each particle, update the personal best position iP and
obtain the global best position gP across the population;
(5) If the stop criterion is met, go to step (7); or else go to step 6
(6) Update the position vector of each particle Go to step (3);
(7) Output the gP as a group of optimized parameters.
Risk Prediction in ALTA

This explains the proposed algorithm for ALTA. The ALTA mainly backed by LS-SVM
regression that relies on IWP-QPSO for feature extraction.
The Development log considered into multitude tuple.

Collect the resultant detailed near features of each tuple
Submit feature matrix as input to LS-SVM regression under IWP-QPSO, which infers
the data to be used for training by generating minimal required number of support
vectors.
Estimate the absolute levels of the features.
Do risk prediction by evaluating the features of the current and subsequent
dependent activities of the code tuple of the current development.
Identify the risk status
Empirical study and results discussion

It is opted to various logs that are of applications of various sizes ranging from
low, mid to large level.
The performance of the ALTA verified by applying on development log of the
Gdownloader, which is tiny open source application, JPCAP, which is tiny open
source application and LUCENE, which is large size open source application.
Empirical analysis is conducted to check the performance of ALTA is as given
below. The development, test phases and their process logs were partitioned into
multitude tuple set.
The extracted features from each process log tuple and generalized with IWPQPSO, then trained with LS-SVM, then attempted to predict the risks in next tuple
of actions.
And then we analyzed performance of the ALTA by comparing the predicted risk
ratio with actual risks of each tuple logged in process log.
So it is concluded that ALTA influences in risk forecasting at Development and

Testing stages.
The figure represents the accuracy of ALTA in risk prediction during development
and testing.
From the results it is clear that ALTA is significant and reliable to forecast defects
during development and testing phases of the software development.
To simplify the results analysis, each application and its log partitioned into
multitude tuple set of 4 for development and 4 for testing. The results description
follows
Table represents the comparison of defects observed during development that are
logged in process logs and predicted at In-Appendage time of development and
testing.
From the results it is clear that ALTA is significant and reliable to forecast defects
during development and testing phases of the software development.
ALTAs performance report on tiny software development:

performance of validation of ALTA on Gdownloader process log
Risk prediction ratio at multitude tuple set of Gdownloader
Development phase:
Defects in
Block 1
Block 2
Block3
Block4
3%
1%
0.30%
30%
3%
0.20%
0.15%
27%
Process
Log
Defects
Predicted by ALTA
Risk prediction ratio at multitude tuple set of Gdownloader

Development phase:
Defects in
Block1
Block2
Block3
Block4
14%
0%
0%
0%
11%
0%
0%
0%
Process
Log
Defects
Predicted by
ALTA
Line chart comparison of defects logged in process log and defects predicted
by ALTA for small size software development
Conclusion
With this we can come to the conclusion that ALTA as in -appendage risk
prediction model can lessen the cost because of defects in development and
testing.
Depending upon the results of the empirical analysis mentioned earlier, it is to be
concluded that irrespective of the size of the software to be developed the ALTA is
best for minimizing cost, using of resources, balancing development and testing
phases.
It enables to work with any of the software development models .and leads to
stable and scalable software development. The methodology used in defect
prediction is stable in all types of software application sizes.

87 PPT

Uploaded by

Copyright:

Available Formats

87 PPT

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

87 PPT

Uploaded by

Copyright:

Available Formats

Measuring Defect Potentials and

Minimizing the Difficulty of SQA

About the work

Metric based models such as maintainability index is an

Prediction models are most advanced form of models

For software engineering, the value-based opinion is commonly considered of

These variations must be accounted for in quality models by defined means of

It should be stated that measurement is important for any control

Hence the measurement of quality attributes is necessary for an

Predictive quality models dont possess definition of the concepts

Many of them depend on regression using a set of software

SQA by defect prediction

DEFECT PREDICTION USING LS-SVM AND IWP-QPSO LS-SVM

According to simplification, Suykens and Vandewalle proposed a modified version of

IWP- QPSO (Intensifying the Worst Particle for QPSO)

The swarm particle can be found using the following.

t1i = (pi - qi ) * f (r)

LS-SVM Regression and QPSO based hyper parameter

y R . In feature space LS-SVM [25] regression model take the form

Hyper-Parameters Selection Based on IWP-QPSO

To solve hyper parameters selection by the proposed IWP-QPSO , every particle

The following steps describe the IWP-QPSO -Trained LSSVM algorithm:

Risk Prediction in ALTA

The Development log considered into multitude tuple.

Empirical study and results discussion

So it is concluded that ALTA influences in risk forecasting at Development and

ALTAs performance report on tiny software development:

Risk prediction ratio at multitude tuple set of Gdownloader

You might also like