"How to write the output equation for random forest and neural net technique?"

vjmeena

Like in linear regression it is possible to write the output equation in terms on y=mx+c. In the same way I want to know how to write the equation for random forest and neural net technique.

rfuentealba

Hi @vjmeena

To my understanding, it is not possible, or very complicated, to express neural networks in terms of mathematical formulas and equations. The most basic way to represent it is: ƒ(x) = x -> y, but since an artificial neural network is a class of algorithms rather than a single do-it-all algorithm, you should define the input, the time, the activation, the threshold, the activation function, the propagation function, the output function, the output and the weight of both.

I'm not aware of any formal definition of random forest, but you might want to take a look at the definition of a decision tree to begin with, as a random forest is indeed a collection of decision trees.

All the best,

Rodrigo.

Telcontar120

Just to add to the good answer already given by my friend @rfuentealba!
There is no ordinary functional equivalent representation for any tree-based model.  A single tree can be represented as a series of sequential if/then statements (rules), but for Random Forest you would have to have a separate ruleset for each individual tree and then another ruleset for aggregating voting across all trees, so it becomes highly impractical to even try to represent this.
There is a functional representation of a neural network model; however the mathematical structure is incredibly complex because of the number of weights and nodes plus the transformations required by the activation function.  You can actually see this output from RapidMiner (check the mod port) but it is tedious to try to represent this elsewhere.
In both cases, complex modern machine learning algorithms like this are designed to have the model scored in the same software that builds and stores the model because there is not a practical way to export the raw model itself for application elsewhere, and RapidMiner is fully capable of scoring any future data that you want to generate a score for.

rfuentealba

Hi @vjmeena

To my understanding, it is not possible, or very complicated, to express neural networks in terms of mathematical formulas and equations. The most basic way to represent it is: ƒ(x) = x -> y, but since an artificial neural network is a class of algorithms rather than a single do-it-all algorithm, you should define the input, the time, the activation, the threshold, the activation function, the propagation function, the output function, the output and the weight of both.

I'm not aware of any formal definition of random forest, but you might want to take a look at the definition of a decision tree to begin with, as a random forest is indeed a collection of decision trees.

All the best,

Rodrigo.

Telcontar120

Just to add to the good answer already given by my friend @rfuentealba!
There is no ordinary functional equivalent representation for any tree-based model.  A single tree can be represented as a series of sequential if/then statements (rules), but for Random Forest you would have to have a separate ruleset for each individual tree and then another ruleset for aggregating voting across all trees, so it becomes highly impractical to even try to represent this.
There is a functional representation of a neural network model; however the mathematical structure is incredibly complex because of the number of weights and nodes plus the transformations required by the activation function.  You can actually see this output from RapidMiner (check the mod port) but it is tedious to try to represent this elsewhere.
In both cases, complex modern machine learning algorithms like this are designed to have the model scored in the same software that builds and stores the model because there is not a practical way to export the raw model itself for application elsewhere, and RapidMiner is fully capable of scoring any future data that you want to generate a score for.