Artificial Neural Network on Software Systems

Actions for Neural Network on Software Systems- Mwalimu

Hello all,

Where:

I would like to use Neural Network method for software or system reliability predictions. Software reliability is critical to business operations and other day-today use by most of us. My personal experience with software help-desk services has enlightened me to a crucial part of supporting software users. Users do not want to deal with consistently failing software. Unfortunately, software is placed on the market for sale even when it may have known issues by the developers. The known issues are not always brought to light by the sales engineers (for an obvious reason), therefore uninformed consumers agree to purchase the software. In summary, during development, the software passes a certain threshold of quality assurance testing and then put on the market for sale. As a member of support services, I have been called upon multiple times to review software quality prior to purchase or implementation. In these cases, I've created summarized assessments that interactively rate the software for decision makers to review and or approve/disapprove. One thing to keep in mind, software may fail because of external factors but ultimately a failure results in a cost to business operations. Predicting software failure could increase profits and improve labor utility.

Who:

A Neural Network predictive method could help improve work-life balance. Help-desk engineers have to be informed on multiple software manuals in order to support their users, and having predictive values could help create critical preventive maintenance plans for a user-friendly environment. Sometimes, software or system downtime's occur unexpectedly, which requires unplanned human labor costs. The cost is not only on the business, but also the help-desk or software administrators from a work-life balance perspective. With preventive maintenance, all users are better off.

Why:

There's no such thing as perfect software. A Neural Network method could support a different functional perspective for analyzing the software by being able to predict its pitfalls and whether or not our company can handle the cost's of software downtime's. Historically, Neural Networks have been applied in a similar manner as explained by R. Gargoor and N. Saleem in their International Journal of Computer Science Issues publication. In their article, they highlight Neural Network methods used to predict specific software failures using software reliability growth models (SRGMs). SRGMs are mathematical models of the software components and infrastructure. Ultimately, Gargoor and Saleem recommend using combined Neural Networks from different techniques for best results; Feed forward neural network, S-shaped model (logistical curve), and GO model (Goel-Okumoto). Though Gargoor and Saleem focused on development software data prior to go-live dates, I would like to focus on post development testing. At this point, users are already utilizing the software and our Neural Network method could  be utilized to create mathematical functions that predict failure in the system. The data could be collected from multiple sources on the server such as the application, security, configuration and system event logs. Actual and predicted results from the Neural Network models by Gargoor and Saleem are shown below from 3 different data sets using a particle swarm optimization technique. Particle swarm optimization is population based stochastic (Random probability distribution) optimization.

References:

http://ijcsi.org/papers/IJCSI-10-4-2-274-281.pdf

By: Mwalimu Phiri

3/19/17