In case anyone is interested, this is what this project turned out to be in the end:
What my program does is the following:
The user trains on a base track (usually consisting of at least half a dozen different corner types etc.) and his fastest lap time is recorded (e.g. from a 15 minute session). The telemetry data from this lap is then compared to a benchmark lap.
Then we use a timing analysis for very short track sectors (e.g. 100m) to identify where the user is losing most of the time. We then decompose the track into corners and straights based on the curvature of the track layout and from that, using the identified problem sectors, construct track segments which contain enough layout context to accurately depict the track segments with which the user is having difficulties.
Based on these segments we then use a shape matching algorithm to identify tracks which contain a high concentration of these segments and suggest to the user to continue training on these in order to focus training on the problem segments. We derived some track suggestion metric which takes into account how much time you are losing in particular segments compared to the maximum time loss encountered as well as how highly concentrated the identified track segments appear in each track that is searched. This process is repeated to keep narrowing down the problem segments until the user reaches a track which consist only of one or two different corner types.
Then the user returns to the base track and does another 15 minute training session in order to assess the training progress made.
At each stage of the program the user is being shown telemetry data comparisons (e.g. plots of the racing line which show braking points, throttle on points, full throttle points, gear selection, rpm, speed etc.) in order to start understanding why he is losing time against the benchmark.
Essentially the aim of this is to make use of the additional freedom offered by virtual simulators as opposed to real life race tracks. In real life, if you are struggling to get this one particular corner right, you will have to keep racing through the entire circuit before you get back to the problem corner. Even through segments that you are perfectly fine with. This is a bit of a waste of time and hence our approach tries to focus this type of training and attempts to allow you to concentrate on particular segments.
Obviously our implementation is far from perfect but I think it is an interesting concept.
Hope you all like it. Thanks again for all you assistance throughout. I have acknowledged this forum in my final report so please believe me when I say that I have truly appreciated the assistance I received on this board.