Hello, i'm working on a systeme that logs events happening inside. errors belongs also to the event. i want to know what are the possible methods that could be used to predict next event/events. basicly, i extracted all possible event that may occure in the system and i created an oriented adjacency graph. from this graph, how can be handled to create (rules of prediction) regards K.Ali
Dear Ali Kaddouri ,
You need to set a probability for each event or (the expected next event). This will help to give weights to the paths when we start at a vertex going toward another vertex.
Then introduce all such probabilities into a weighted matrix that represents your mathematical model as a directed graph.
Finally, follow Markov chains style to compute probabilities that help to predict the next event after a ( finite or infinite) number of steps ( or actions based on your model).
Best wishes
This probability was already calculated and set based on a banch of events occurence in the system. i know there are markov chain concept that could be used by i want more method which could be used in my research
Luis Gomez , Could you please give me an overview on the steps to follow so i can follow in my research.
thanks
It's not really possible to give a good advice without looking at the data and not knowing the problem, but you have 2 ways: classification or regression. According on how you want to approach your problem, there are different kinds of ann or machine learning tools to use. For instance, if you can define a finite amount of events to predict, you can use a classification ann or a ML tool like a SVM, otherwise it might be better a regression ann. Then you have to analyze and decide if your data call for direct input into the model, or you will need some kind of feature extraction from them. Finally you have to give a decent amount of training examples to the model, check the results and modify the parameters and repeat in the case that the results aren't satisfactory. There are probably other methods or tools, like statistical ones, but they are not my expertise.
One way to easily introduce into this subject is trough matlab deep learning and statistic and machine learning toolboxes.
Thanks Luis Gomez, i will try to look inside each method you quoted in your reply and see what it will give us.
- Badges
- Science topic
- Similar topics
- Phytochemistry
- Extracts