How to improve classification prediction models accuracy?

Research & Models
,
#1

Hi,
I have applied several classification methods, unfortunately, the developed models never exceed 62% of accuracy.

here I attached a comparison table of the developed models.
I’m wondering how I can improve the models’ accuracy!?

Accurcy Table
Accurcy Table699×451 17.1 KB

#2

The first question i suggest to you is:

  • what are your performances on the training set?
#3

I have split the data into training and testing and the confusion matrix give these results, not sure if is it the right thing

from sklearn.tree import DecisionTreeClassifier
dt=DecisionTreeClassifier()

dt.fit(X_train,y_train)

pred_dt_tr=dt.predict(X_train)

pred_dt=dt.predict(X_test)

from sklearn.metrics import confusion_matrix,classification_report,f1_score

print(confusion_matrix(y_test,pred_dt))

print(classification_report(y_test,pred_dt))

#4

What I meant, I suppose that the table is from the testset.

So what are the models performances on the training set?

This could be a useful starting point to understand if:

  • You still have a margin to learn with the current data

  • You have generalization issues or you model is overfitting

  • Your model capacity is limited

  • Missing hyperparameter tuning on a validation set

  • Etc.

#5

thank you for replying
How can I check the training performance?

#6

I just see the accuracy for training and resting for KNN. the accuracy for training is 0.996 and the testing is 0.722

#7

Using pred_dt_tr and y_train

This forum is generally about Tensorflow but you are using sklearn so I suggest you to use sklearn support channel for sklearn code/projects.

I don’t know your specific learning goal and dataset but in TF you can try to explore:

1 Like
#8

Sorry I apologies if posted something not related to tensorflow policy.

Thank you for replying to me

#9

No prob. Let us know If you have other questions experimenting TF.

#10

It is almost impossible to suggest anything without additional information.

  • How many classes are you trying to predict?
  • How many training images do you have?
  • Does your dataset suffer from data imbalance?
  • Have you already checked your data? (e.g. if it is labeled correctly)
#11

I’m trying confirmed and suspected cases
it is numerical data not images
How can I check for data balance?
the data is labeled

#12

It would be great that you will show two graphics: acc & loss, as a picture …they can be drawn on one picture. It has given some answers to the appeared questions.

#13

unfortunately I only applied Confusion Metrix so I didn’t apply anything else