0

I work for a manufacturing company that collects times (man-hours) for tasks. The tasks are similar in concept from one product to another, but the characteristics of the product can vary (like size, voltage, etc).

It has been something like ten years since I've had to do any heavy lifting in terms of data analysis... I'm trying to take our dataset (about four years' worth of data) coupled with some parameters about each of the products and use it to ultimately be able to input a product's characteristics and be able to predict the output in hours.

Can anyone help me get started with this? I'm an EE by background and have worked with neural networks and linear regression in the past, but it's been something like 10 years, so I've forgotten it all.

Thanks so very much, Timothy

Griphus
  • 1
  • A little more information about the data you have would be welcome. For instance, your response to @IrishStat's answer suggests your data might be time series, but nothing in this question suggests that this is the case--and in fact it is difficult to see how such data could usefully be construed as, or modeled as, time series at all. It would be especially helpful to have some clearer idea of what a product's "output" is and how you expect that to be related to the other characteristics. – whuber Aug 09 '14 at 19:01
  • Hi, my dataset will include man-hours charged to specific operations each day over a course of several years. The operations have the same name from product to product (for example, one of them is "connect"). However, the amount of time "connecting" takes varies dramatically from one product to another. The thought is that I can do some sort analysis comparing time to connect vs. some characteristics of the product (like how many connections), and find the most critical ones as well as predict/estimate times for future product builds. Let me know if that doesn't address your comment. – Griphus Aug 11 '14 at 02:53

1 Answers1

0

The method is called Transfer Function or Dynamic Regression. It is a super-set of linear regression that takes into account leads and lags of the user-specified predictor series. It also allows the incorporation of lagged effects of the error term. Additionally it allows the identification and incorporation of level shift/seasonal pulses and local time trends. Furthermore one time pulses are adjusted for. All of this while considering parameters and error variance that possibly change over time. A good on line reference for introductory time series material is available at http://www.autobox.com/cms/index.php/afs-university/intro-to-forecasting.

IrishStat
  • 29,661