Based on recent experiences, I have started to realize that simpler predictive models, a simple CNN model architecture or ANNs model, could yield good results, if not better that more complex models.
If you had faced a timed challenge to build a predictive model and you found peers building complex models using feature extractors, learning rates, or other methods, would you be going to try a simple CNN model? and would you be optimistic that you could achieve results better than others for the challenge?
Thank you!
Dear Mr. Morani!
You raised an important issue. A simple approach might yield better results than complex models. Predictive COVID-19- research can be a case -study to consider, the outcome is case - and context dependent one:
1) Atchadé, M.N., Sokadjo, Y.M., Moussa, A.D. et al. Cross-Validation Comparison of COVID-19 Forecast Models. SN COMPUT. SCI. 2, 296 (2021). https://doi.org/10.1007/s42979-021-00699-1 Free access: Article Cross-Validation Comparison of COVID-19 Forecast Models
2) Jimenez-Solem, E., Petersen, T.S., Hansen, C. et al. Developing and validating COVID-19 adverse outcome risk prediction models from a bi-national European cohort of 5594 patients. Sci Rep 11, 3246 (2021). https://doi.org/10.1038/s41598-021-81844-x Open access: https://www.nature.com/articles/s41598-021-81844-x
3) Weili Kong et al. (2021). Prognostic value of three rapid scoring scales and combined predictors for the assessment of patients with coronavirus disease 2019, Nursing Open, Early View, 03 June 2021, Open access: Article Prognostic value of three rapid scoring scales and combined ...
Yours sincerely, Bulcsu Szekely
Kenan Morani I agree with Bulcsu Szekely that your question is intriguing.
- First, it presents one of the most crucial yet swept under the rug issues of the computer age, is modeling part of the scientific method?
- Two, unquestionably, simulation yields practical real-world results, anyone who looks at the data on crash testing or computer modeling of airplane design can not help but be impressed
- Three, however, I as you seem to suggest that the more complex a model is, evidently in order to closer match reality, the less efficient or even less reliable its results may become.
- Finally, when you introduce interrelated functions, where the calculations of one affect the other's output, you run into Chaos in the form of of the Three Body Problem, once you have included three interrelated functions. Unlike two-body problems, no general closed-form solution exists, as the resulting dynamical system is chaotic for most initial conditions, Thus numerical methods are generally employed. Of course, those methods are approximations
Thanks for an excellent question
CGY
Dear researcher, I totally agree, simple predictive models are very useful and easy to optimize. I think that more complex models are more useful in more complex contexts, more complex processes. The critical diagnosis of the situation under study is essential to determine what we can select in each case. Thank you
Raúl Rodríguez Muñoz in general what you say makes intuitive sense, simple models for simple problems, complex simulation for sophisticate cases. But I cannot think of a mechanism that would back you up or justify your statement. Could you share your basis for why you think that? General rules in science seem to be a lost art form. If you have I glimmer of an idea as to why that would be so, then it would be a very important paper to write. Tnx You
CGY
It might be that when we deal with sufficiently big datasets then hand-engineering designs become less important than focusing on the learning of the machine learning model.
- Badges
- Science topic
- Similar topics
- Psychology
- Mental Processes
- Learning