iSixSigma

Defect detection models

Six Sigma – iSixSigma Forums Old Forums Software/IT Defect detection models

Viewing 5 posts - 1 through 5 (of 5 total)
  • Author
    Posts
  • #26171

    Tierradentro
    Participant

    Hi all,
    Our organisation is looking to establish Statistical Process Control for the weekly rate of defect detection during our Build and Test Phase. I recently found Putnam’s suggested use of the Rayleigh curve for estimating the number of defects during the lifecycle; it looks like a decent fit for what we want.
    Before I go ahead and invest too much time,  has anyone had any practical experience with this model ? If so, I would appreciate any advice/pointers/feedback.
    Thanks in advance for your help. With regards,

    John
     

    0
    #63641

    Gary A. Gack
    Participant

    hi John,
    We have seen a number of organizations using Rayleigh successfully, with a couple of important qualifications: (1) there will always be an error band around the prediction, so don’t take the forecast too literally (2) the prediction presumes a reasonably consistent level of defect detection effort during each time period – if that varies significantly so will the defects discovered (3) use it to forecast defect discovery for each specific type of detection activity – e.g., functional testing will have one curve, stress testing another curve that may overlap, but show a similar pattern – an overall curve that summarozes the smaller curves can also be useful, but probably has a larger error band so is less sensitive (4) as you probably know, Rayleigh only forecasts the distribution of discovery over time, given a forecast of total defects (it does not forecast the expected total – that needs to be developed from history data and/or from an estimating tool such as Putnam’s [qsm.com] or SPR’s [spr.com]).
    good luck – let us hear how it works for you if you decide to use it

    0
    #63642

    Tierradentro
    Participant

    Hi Gary
    Thanks very much indeed for the feedback. Sounds promising.
    > (1) there will always be an error band around the prediction
    This is a very good point and has bothered me since we considered adopting Rayleigh.
    I’ve been trying a bit of analysis based on some recent projects. Overall, the Rayleigh curve was a good fit (surprisingly so !) – but there were instances where we had discovered (eg) 2x the number of faults that the model predicted.
    How would we establish proper Statistical Process Control using such a model ?
    I guess that I need to *fully* analyse our fault discovery process to determine if it is truly stable (ie, taking into account special causes & removing them) to see if Rayleigh is truly appropriate.   Suggestions more than welcome !
    Again, thanks for all your help. 
    Regards

    John

    0
    #63662

    bkshah
    Participant

    In CMMI, there is process areas call OPP. This PA defines clearly model for defect detection. You need to findout defect analysis based on past closed project(dev and maintenance). Based on this you need to define baseline for organization . Later on this can be implemented in the project
    B K Shah

    0
    #64808

    chhabra
    Participant

    Hi,
    Actually I also have a query on the defect prediction itself ,Say my Org as of now not having any prediction model Can I use the industry Defect distribution over the SDLC phases,Could yuo please let know that If my Org defect data does not follow the Rayleigh Curve even then Can I use the Rayleigh model or If not Could you please syuggest me some otther defect prediction model,OR could you please let know the Bases of the Rayleigh Model

    0
Viewing 5 posts - 1 through 5 (of 5 total)

The forum ‘Software/IT’ is closed to new topics and replies.