# How to build a predictive model using sound (frequency and amplitude) as the input

I'm working on a medical research project where we're looking at the changing resonances in femur bones during hip replacement operations. Some surgeons use the increasing pitch of the bone resonance to tell them when to stop hammering/reaming the shaft of the feamur. If they apply too much pressure when creating the cavity for the replacement hip, they risk fracturing the femur.

Basically, I've recorded a number of these changing resonances using a bone conduction mic and now I have a number spreadsheets with one column listing frequencies in 5Hz bands and 3 columns showing the amplitude of that frequency when the bone is loose, medium to tight (referring to the tension of the reamer/chisel creating a cavity for the replacement joint).

I'm coming at this more from a sound perspective so I would really appreciate some advice from anyone with a handle on stats and predictive modelling, I'm trying to find a pattern to the change in frequency which is consistent in all cases.

Any advice would be very much appreciated.

-
A (small) example of your data would be helpful, for otherwise you can only get generic advice. –  whuber Aug 9 '12 at 13:35
What exactly are you wanting to predict, whether or not the bone breaks, whether it's loose, medium, or tight, or the amplitude of the frequency? Furthermore, you ask about predictive modeling, are you actually going to use this model to make predictions in the future to guess at the output value when you have the input values, or do you want to understand why a given result happens? –  gung Aug 9 '12 at 13:44
hi whuber and gung, many thanks for your reply. here's a link to one of sequences of impacts (loose to tight), NB the way tyne resonant sound changes; soundcloud.com/jack-bobson/10a-reamer-sequence-loose-to –  Bob Jackson Aug 15 '12 at 10:31
ultimately what we'd like to do is use realtime analysis of the changing resonance to tell the surgeon when to stop chiseling/reaming to avoid fracture. so, first of all we need to analyze the changing frequency of a series of recordings, try to identify if there's a consistent pattern. stats are not my strong point so I'm looking for some help in analyzing results from an excel breakdown matlab analysis; here's an example: fileden.com/files/2012/8/14/3336912/10b.xls. I have around 15 similar excel files from 15 different recordings with reamers of various sizes. any ideas? Bob –  Bob Jackson Aug 15 '12 at 10:50