Hfrxdjt

Two training sets are involved, one complete, one with missing feature data as well. The data consists of CNNs and GIST features.

For the normalising, I have MinMax Scaler feature. I have cleaned up the missing data by using the mean of the column because I tried the mean by row but this bring down the accuracy of the classifier further down. I am assuming that this because the average of all the features for the specific photo doesn't calculate well.

I then concatenated both the datasets. Is calling the fit method twice incrementally better?

Classifier Results

• low accuracy (70%)


• 
  

  Log loss is 9

  
  
  

   precision recall f1-score support
  
   0.0 0.67 0.56 0.61 431
   1.0 0.72 0.80 0.76 605
  
   micro avg 0.70 0.70 0.70 1036
   macro avg 0.69 0.68 0.68 1036
   weighted avg 0.70 0.70 0.70 1036
  

  
  

I also have tried multiple train-test splits, I achieve the best accuracy at 0.6 train.
I understand this is a broad question.
I have tried both logistic regression with saga and liblinear. SVM with rbf too. But still unable to increase the accuracy of my classifier.

I plotted my training set data of one feature from both classes and the data appears to be non linearly separable? As in the data from 1 and the data point from 2 appears to be all over. I am not sure how else I can do this?

Also how can I attach confidence of the training data into my classifier? As In I have the confidence for each record of data. ID 1 - 0.2, ID 2 - 0.4 and so on..

I am new to the subject, apologies if any of it sounds dumb.