Neural Networks versus Random Forest
Neural Networks and Random Forest are being used both for Classification and Regression related problems in Machine Learning.
In this post let us briefly discuss both these algorithms in terms of Performance,Cost & Time,Robustness.
Before that let us quickly touch upon briefly about these two approaches
Random Forest : It is an ensemble of Decision Trees where the leaf node will be majority class for classification or average for regression problems.Random Forest is made up many Decision Trees, the logic is together we win.Random Forest avoids overfitting because of this amalgamation of smaller trees.
Neural Networks : They mimic human brain replicating similar kind of behaviour.Just like brain it is modelled as many small nodes that gets activated based on the weights ,bias and activation functions.
Performance : Performance here means quality of classification or regression.Both the algorithms has potential to handle complex nonlinear relationships.Due to construction Neural Networks has an edge over the Random Forest.
Cost and Time: RandomForest requires less inputs for building the model and they can handle both numerical and categorical features well.There is no need of feature normalisation .On the other hand Neural Networks requires large amount of data and they continuously improve on accuracy.
Neural Networks require GPU enabled machines which costs higher.The Trail and Error for Neural Network is also a time consuming operation.
Comprehension: In many places such as Finance a loan may or not be offered, a credit card fraud happens.Humans needs to understand and interpret the reasons for not offering loans or fraud.Neural Networks with lot of parameters would make life difficult to comprehend this data.
Random Forest also have similar issues but recently some approaches has been developed to identify the representative trees.
Unstructured Data: Random Forest is good when data is structured whereas Neural Networks can handle unstructured data.
Use Neural Networks
Create a model where there is NO need for identifying the variables at play
Models that are dealing with images,videos,text classifications
Use Random Forest
Models that needs comprehensions
Need low cost training