A general Question in Machine Learning

#1

I have completed almost the entire course. While solving the challenges I always had this doubt.
How should I improve model performance.?

In the leaderboard, many students have got high accuracy than me.
For eg. - In the titanic challenge, house price prediction, movie rating.
How other students manage to get high accuracy. How do they approach any ML challenge?
Even I tried changing the model (different algorithms) still there was no significant improvement.

Last week I was doing ZS hackathon on machine learning binary classification problem. During the first 5 hours, I scored 0.867 score, but for the next 36 hours, I could not improve this score, while others have scored 0.97.

So, Initially, when I have built one model, How should I focus on improving its accuracy.? Building a model is easy, but improving it, is a real struggle.

I would like you guys to guide me on this.
@rachitbansal2500 @Rahul_garg @Manu-Pillai-1566551720093198 @yash97

1 Like
#2

I also have same question …

#3

hi @mohituniyal2010
i was also attempting the challenge
the challenge was more about data analyst placing the missing data
i got 93.5% accuracy in the challenge
i used binary classification using neural net double dense layer
after analysing the data i came to conclusion the data was removed delbrately from test cases
so using imputer class i filled the data but that didnt increasesd my accuracy
i also one hot encocded inputs and using k fold cross validation

#4

Did you consider all the features? bcoz many of them were redundant like remaining-time.1 .
and most of the features had high missing values, i tried filling values sometimes with " ffill ", mode, and median as well…
I can’t understand what more i could to do to increase the accuracy…

I also did one hot encoding, tried normalising the data, tried neural networks, tried dropping some not correlated features with y_label, tried to take only most important features from tree.feature_importances_, and many different approches.

Finally I used ensemble with xgboost and random forest. Still the accuracy didnt improve :laughing:

#5

yes i dint use all features
the data they provided was not featured enginnered correctly none any documenttion about describing data was specified

#6

I hope I’ve cleared your doubt. I ask you to please rate your experience here
Your feedback is very important. It helps us improve our platform and hence provide you
the learning experience you deserve.

On the off chance, you still have some questions or not find the answers satisfactory, you may reopen
the doubt.

#7

Regulated AI calculations can apply what has been realized in the past to new information utilizing marked guides to anticipate future occasions. Beginning from the investigation of a known preparing dataset, the learning calculation creates a surmised capacity to make forecasts about the yield esteems. The framework can give focuses to any new contribution after adequate preparing. The learning calculation can likewise contrast its yield and the right, planned yield and discover blunders to adjust the model as needs be.

Source : Machine Learning Course in Pune