Dismiss
InnovationQ will be updated on Sunday, Oct. 22, from 10am ET - noon. You may experience brief service interruptions during that time.
Browse Prior Art Database

A prediction method for post shipment defects in a product

IP.com Disclosure Number: IPCOM000199888D
Publication Date: 2010-Sep-20
Document File: 2 page(s) / 63K

Publishing Venue

The IP.com Prior Art Database

Abstract

This article presents an equation which can be used for the prediction of the number post-shipment defects which will be found by users of a software product. The equation contains terms which capture the main factors influencing both the injection and discovery of defects.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 55% of the total text.

Page 1 of 2

A prediction method for post shipment defects in a product

Disclosed is an equation which can be used for the prediction of the number of post-shipment defects (commonly and hereinafter referred to simply as defects) which will be raised by users of a new release of a product during the first 4 years after the product is released for sale. This type of prediction is a common requirement for service teams and the empirical equation disclosed provides a useful method of making the prediction.

    A common problem posed to service teams is to predict the number of defects likely to be raised on a new release of a product. This information is important in projecting personnel levels required in service teams or highlighting where recruitment and/or increase in skills will be required.

    A typical solution is to use the previous year's defect receipts and "challenge the team" by saying that the prediction is 10% less than this. Clearly, this is not a true prediction, merely a desired target.

    Another solution which is sometimes used is to multiply the shipped number of KLOC (1 KLOC = one thousand lines of code) figure for a product by a standard number and call that the prediction. Again, this is not a particularly good prediction algorithm since many important factors are not incorporated into the prediction.

The proposal for this disclosure is an empirical equation which can be used to make better predictions.

The equation is
A = Q (N z / S) (xB + yD)

    The variables are as follows. A is the total number of defects expected in the four years post-shipment of a product. B is the number of KLOC of base code (code which has not been changed since the previous release of the product) and D is the KLOC of changed code in the new release. x and y are multiplication factors. N is the number of licences expected during the four years and z is another constant to be determined. Q is a quality of test factor.

    This equation codifies all the main factors which affect the defects to be expected. If testing of a particular release has not gone well, Q, which is normally unity, can be increased. N can be assigned based on marketing data or other knowledge of likely customer uptake of a release. x and y can be deduced by fitting the equation to data from previous releases of the product. Having separate factors for base and changed code allows for large and small releases to give different defect projections and allows for differential defect probabilities in base and changed code. z is a fr...