Support vector machines

Discussion in 'Big Data and Analytics' started by Narayana Surya, Dec 23, 2018.

  1. Narayana Surya

    Narayana Surya Well-Known Member
    Alumni

    Joined:
    Feb 27, 2018
    Messages:
    59
    Likes Received:
    0
    HI
    Q1. Can you please explain me how SVM will do classification if i have more then 2 features in my data set..?(AS per my understanding each record of feature will be assign in each coordinate )
    EG:
    i have features as SAL,EXP,AGE,RAT and i want to classify their position depending on this like(Associate,Senior Associate...etc) i want to know how it will classify by considering all these features

    Q2. How can i classify items into multiple categories in SVM (generally we will classify it into 2 categories)

    Q3. How can i introducing additional feature in SVM (i know we can do it by using kernel function but some times formula are using so i just want to know difference b/w formulas and kernel types(Linear,polynomial)...?
     
    #1
  2. Vikas Kumar_18

    Vikas Kumar_18 Well-Known Member
    Simplilearn Support Alumni

    Joined:
    Dec 17, 2018
    Messages:
    177
    Likes Received:
    31
    Hi,

    Q1: Ans:
    We can easily visualize the 2 dimensional space but if we have 3 or more dimensional space then we cannot easily visualize and will go for the mathematics usually matrices to represent. SVM always takes the coordinate using all the features and draw that particular records.

    Just check the below mentioned url for 3 dimensions, you can visualize like this:

    https://www.google.com/search?q=3 dimensional coordinates&rlz=1C1GCEU_enIN827IN827&source=lnms&tbm=isch&sa=X&ved=0ahUKEwjhlrG2w7jfAhUIdysKHY05ANEQ_AUIDigB&biw=1600&bih=758#imgrc=jwdbEqaS-0KIxM:

    Similarly, coordinates could be made using 4 variables and visualized as:

    https://www.google.com/search?rlz=1C1GCEU_enIN827IN827&biw=1600&bih=709&tbm=isch&sa=1&ei=494gXLuOJor8rQHquqbwCQ&q=4 dimensional coordinates&oq=4 dimensional coordinates&gs_l=img.3..0.74481.74983..75199...0.0..0.95.184.2......1....1..gws-wiz-img.......0i7i30j0i8i30.YlEyXUlQDS8#imgrc=IDgHqCndKBknVM:

    But it would be difficult for us to understand but machine can understand. If we want to understand then go for the matrices representation like this (SAL,EXP,AGE,RAT).
    Q2: Ans:
    SVM works as "One vs One" in case of binomial categories and "One vs all" for multinomial classification. Let's assume we have 3 classes Associate, Senior Associate, Junior Associate. firstly, it takes the Associate as one class and rest other (Senior Associate, Junior Associate) as second class and train model and then Senior associate as one class and others(Associate,Junior Associate) as second class and so that it trains the model.

    Q3. Ans:
    Usually in SVM linear and Kernel options deals with the linear data and nonlinear data. When our data is linear then just go for the linear SVM and it works fine but if it is nonlinear then various kernels are used to make it as as linear then do the classification. Kernels refers as hyperplane (multidimension).
    Please understand as if any data is nonlinear in N dimensions then if we would like to make it as linear then we need to project that data into (N+m) dimensions where m =1,2,3,4... and there is no hard and fast rule it's on trial and test the results.
     
    #2
    Last edited: Dec 24, 2018
  3. Narayana Surya

    Narayana Surya Well-Known Member
    Alumni

    Joined:
    Feb 27, 2018
    Messages:
    59
    Likes Received:
    0
    Can you please give me small example how non linear data is converted to linear by using kernel ( i am having doubt how non linear data is converting into linear data )?
     
    #3
  4. Vikas Kumar_18

    Vikas Kumar_18 Well-Known Member
    Simplilearn Support Alumni

    Joined:
    Dec 17, 2018
    Messages:
    177
    Likes Received:
    31

Share This Page