Regarding the value of K in KNN algorithm

#1

as we increase the value of K then the accuracy of the predicted value becomes more am i correct??

#2

hey @amankharb ,
Its not necessary that your accuracy will always increase if you increase your value of K.
Value of K usually depends upon the type of data , and you have to find out the right value of K by taking random tries on its value by either increasing or decreasing it to get better results.

#3

yes got it i was trying this thing only but it will be better if k is odd

#4

if i choose differeng values of k and at a particular value of k if my training predictions are almost equal to Y then that value of k will perform good on x_test also??

#5

i can reach only 74%accuracy in Diabetes Detection Challenge with the help of changing value ok K in KNN and everytime i change value of k the accuracy ranges between 68 to 74% does that mean it cant be more accurate than 74%??

#6

yeah you should always try with odd value of k only .

#7

Not Necessary , sometimes it may lead to overfitting .
You need to check that your accuracy score on train_set should be almost equal to accuracy score on test_set. This is the basic thing we check to understand the way our model is performing.

i can reach only 74%accuracy in Diabetes Detection Challenge with the help of changing value ok K in KNN and everytime i change value of k the accuracy ranges between 68 to 74% does that mean it cant be more accurate than 74%??

Yeah you are right with KNN, if you try tuning its parameter values then too , you can reach your score to 76-77% maximum , not more than that. So yeah you are right with KNN you are limited between this score range only.

Try different other models to further improve your accuracy.

#8

Hey @amankharb ,
If your doubt is resolved , then i would request you kindly mark it as resolved and also provide your valuable feedback as it helps us to improve this platform and provide you with better learning experience.

Happy Learning :slight_smile:.