Machine learning projects have different risks than conventional programming. While our past experience, productised tools and re-use of existing machine learning models defers some of the risk, there’s always some degree of experimentation and risk:
- Feasibility – The problem you wish to solve might not be solvable using sensor data.
- Accuracy – While the problem might be solvable, the resultant accuracy might not be good enough or the development activities might end up taking too long to achieve a required accuracy. You will never get 100% accuracy and if you need this, machine learning isn’t the solution. Before you hastily discard machine learning, remember that many non-machine learning and manual safety critical processes usually have some possibility of failing.
- Performance – If the input data is very large, it might not be possible to derive a model in a reasonable elapsed time. If a model can be created, it might not be fast enough during inference (real use).