Points tomer said

[yiftachn]

1. Check if naive bayes performance decrease on different sample on the data.
2. What is C in the SVM/SVC - understand it deeper and try smaller than 1 C
3. How the model behave to the standartize of making it binary. or binning it by percentages (but dont do percentage over the zero). it should also help NB.
4. Desicion Tree - Check how well is the perfomance.
5. Calculate the Training Accuracy.
6. Data exploration - Feature selection - what single features makes the best model? sequentional feature selection (greedy algorithm).
7. False analysis the model - what charecterizes the False Positive? and the False Negative?

[yiftachn]

Hard negative mining - what charecterizes the example you are most wrong on. and than in iterative training you can show those examples to the model more times.