We need help with Development of a Machine learning algorithm and predictor. We have built a special measurement device and collect measurement data from material probes. The target value is the strain in the probes which is measured with another reference method.
So we have the measurements and the target value in one table. Because we could not fit the measurement to the target by any regression or physical correlation we built a quite good Random Forest with R to predict the target from measurement data. The Forest is good but not good enough.
Your task is to find a better predictor with a better fitting than the Random Forest.
1. Understand the problem and the Input table.
2. Suggest one or more better machine learning algorithms than the Random Forest.
3. Build the model with learner and predictor.
4. Cross Validate your model. Predict the target from measurement data.
5. Test the model on unknown probes.
6. Decide for the best model.
7. Finalice the model for our use.
8. The model must be made in a way that new measurement data can be included (the training data set grows over time), the model is trained again, the new trained model shall cross validate with the hole dataset if the prediction gets better or worse than before.
We would like to use one of the new deep learning algorithms like tensorflow, Keras etc. The tools used should be open source. It would be also nice if your model can be testet in Knime or Weka via a GUI interface.
If you are interested we can send you the dataset. It is not very large but it needs some explanations.
I am willing to pay higher rates for the most experienced freelancers