| Abstract |
Finding, locating, and resolving software defects takes a lot of time and effort on
the part of software engineers. Manual testing are required to search and analyses data in
traditional testing. Manual testing are prone to making incorrect assumptions, resulting in
distorted results, which leads to defects being undetected. Machine learning enables systems
to learn and use what they have learnt in the future, providing software testers with more
accurate information. Several advanced machine learning approaches, such as deep learning,
are capable of performing a variety of software engineering tasks, including code completion,
defect prediction, bug localization, clone detection, code search, and learning API sequences.
One of the most essential methods of examining software quality assurance is software testing.
This procedure is time-consuming and costly, accounting for over half of the total cost of
software development. We're looking to use automated methods to reduce the cost and time
of the test, in addition to the cost issue. The nature of the test, which looks for errors in the
programs, is such that software engineers aren't interested in doing it, so we're looking to use
automated methods to reduce the cost and time of the test. We conducted a survey with
comparison between Machine Learning, Artificial Intelligence and Data Mining algorithms
that can used in Software Testing such as: Hill-Climbing Algorithm (HCA), Artificial Bee
Colony Algorithm (ABC), Firefly Algorithm (FA), Particle Swarm Optimization (PSO),
Artificial Bee Colony Algorithm (ABC), Genetic Algorithm (GA), Ant Colony Optimization
II(ACO), Artificial Neural Network (ANN), Support Vector Machine (SVM) and Hybrid
Algorithms.
All the previous research talked about algorithms without using dataset or found better
accuracy between them. This paper presents a comparisons of classification algorithms such
as: Logistic Regression, K_Neighbors, Decision Tree, Random Forest, Gradient Boosting,
AdaBoost, Linear Discriminant Analysis, Quadratic Discriminant Analysis, Gaussian NB,
Support Vector Machine and deep learning algorithms such as: Artificial Neural Network,
Convolutional Neural Network. We have employed PROMISE (CM1: 498 observations,
KC1: 2109 observations, KC2: 520 observations, PC1: 1109 observations) data from the
directory of NASA to conduct our research. We split the dataset into training dataset (70%)
and testing datasets (30%). We brought up a machine learning hybrid approach by Particle
Swarm Optimization (PSO) and Artificial neural networks (ANN) to overcome the ongoing
problem. The proposed model also presents higher evaluation in the terms of other criteria,
reduce time and space complexities.
The experimental results revealed that PSO-ANN to overcome the problem of the curse of
dimensionality and reduce the computational requirements of the proposed task. It was the
major problem with the previous research methodology mentioned then after we used PSO
with multi-layer (ANN), which is a very powerful methodology of classification in ML. we
have found better accuracy (CM1: 91.0%, KC1: 86.0%, PC1:94.0%) than other methods. |
