Debugging Bad Results

The process of building a good model may involve some iterations of trial and error. If you are not getting good results in a model it’s good to check the following things to improve your results.

  • Collect more data – If you are working with a small data set, it’s likely you do not have enough data to find a good match.

  • Check your sensor placement – Is your sensor oriented consistently throughout your training data? A bad sensor placement can cause very unpredictable results

  • Evaluate your labels are consistent and accurate – Are your labels consistent throughout your project? Check spacing and start/end placement to verify you are labeling your events in a consistent way.

  • Evaluate different algorithms, seeds, segmenters – You may need to try out different parameters in the model building step above or move to the Advanced Model Building step below to get better results if your events are very complicated