I am looking to use different machine learning approaches to build some regression models. I am not sure if it will possible to use machine learning methods such as SVM, LDA, kNN, NB, etc for regression purposes? with limited data?
Overall, what are your recommendations?
Yes, you can depend on how to want to use it you can alter the last parts of the algorithm. For example:
SVM : https://www.analyticsvidhya.com/blog/2020/03/support-vector-regression-tutorial-for-machine-learning/
LDA: https://stats.stackexchange.com/questions/31459/what-is-the-relationship-between-regression-and-linear-discriminant-analysis-ld (Really good discussion)
KNN: https://www.analyticsvidhya.com/blog/2018/08/k-nearest-neighbor-introduction-regression-python/#:~:text=As%20we%20saw%20above%2C%20KNN,points%20in%20the%20training%20set.
NB: https://link.springer.com/content/pdf/10.1023%2FA%3A1007670802811.pdf
Yes. See the link for general understanding and see the paper also.
1. https://www.javatpoint.com/regression-analysis-in-machine-learning
2. https://www.tandfonline.com/doi/pdf/10.1080/09599916.2019.1587489?casa_token=owUFAAF3OR8AAAAA:y4KpnhlHPalG680To9h-hHhMaf17XS2E3iTWQWKTBIdZ7LT0cwltXjy9x7pmcxPWh4CZlYhPUQvo
Safial Islam Ayon Thank you for the links provided. I will go through the links and read the paper.
Really appreciate your answer.
It depends on how limited the limited data is. Yes, it is possible to build regression ML models with a few data. The risk you run is that of highly biased (underfitting) results. The best way to know if you have a good sample size is to test out the ML model on your data in incremental order, and see what your error or accuracy tr end looks like.
Akinbode A. Adedeji, thank you for the clarification. That was the problem of knowing whether you have enough data for ML or not? I am going to start with simple regression models such as PLSR, OLSR, etc to see the performance.
"The best way to know if you have a good sample size is to test out the ML model on your data in incremental order, and see what your error or accuracy tr end looks like." This is a great point to consider. I will test out the ML model/s based on this and evaluate the results.
Thank you for the answer.
Yes, you can use ML algorithm for regression although you have limited data, see the attached reference for the application of ML for regression. Also, you can use Deep learing (DL) methods like LSTM if you have more than 2000 data points. DL performs well than ML. see the reference link. Hope these will be useful.
Best of Luck
Lakmini
Reference link:
Article Drought forecasting in eastern Australia using multivariate ...
Article Deep Learning Neural Networks Trained with MODIS Satellite-D...
Yes, however., during model selections, you have to decide weather the data is numerical or non-numerical? Regression is also classification, however., it is for the examples of numerical data.
A data-driven model constructed with limited samples, can not be considered significant. Try to use as much data as possible. Also, use sufficient input parameters.
I recommend you go through the following paper as it has useful information that would answer your question:
Article Classification of Indoor Environments for IoT Applications: ...
Alfadhl Yahya Alkhaled
check out some sessions:
https://www.youtube.com/watch?v=Oe-h3g0uijI&t=1869s
https://www.youtube.com/watch?v=vphcUWoePm0&t=1999s
and
also some of the articles:
Preprint Statistical analysis of national & municipal corporation lev...
Article A comparative study of modified SIR and logistic predictors ...
yes, there are regressor versions available for mentioned machine learning algorithms but the efficiency of each developed model depends upon data. all these algorithms work well with small to medium sized dataset, however, data behaviour is important while choosing any algorithm.
note: sklearn provides such functionality.
- Badges
- Science topic