AI-based techniques such as DNN, Fuzzy, SVM, and ANN show different forecasting performances for different pollutants such as Particulate matter,SO2, NO2.
Say for e.g. - DNN outperforms other techniques in PM forecasting
- SVM outperforms other techniques in SO2 forecasting.
I am curious to know about the deep logic behind this response of AI techniques.
P.S.- I know the source and origin of the air pollutants might be having some important role to play in this matter.
Dear Adil,
Since you included SVM then you mean supervised machine learning algorithms (MLA). Some authors describe MLA as a subset of Artificial Neural Network (ANN). There are several reasons which make one algorithm better than the others in forecasting pollutants:
1- the size of the training set (the more training data are provided the better is the algorithm). In case the training set is small in this case some algorithms work better than others.
2- Type of kernel used with SVM as an example (Linear, polynomial, RBF...etc)
3- Size of the ANN network (number of neurons, number of nodes, number of layers)
4- Initial parameters setup such as % of validation, training, and test data
In general, ANN is a black box we can only see the result and base on this we can change the settings.
Dear Adil Masood ,
Different methods considers different algorithms and implementations, for example DNN are good for Image data, SVM in some other points and so on, and for DNN you need to tune a lot of Hyper parameters as well like other methods, but your noise or missing values also could impact on your solutions which some algorithms are much more sensitive to noise than others.
Because the pollutants chemical and physical composition would not be same. They never behave same in nature. Same would be predicted by AI.
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Artificial Intelligence