if i have 1200 instances total and i want to classify in binary class. One class is "yes" having 900 and second class is "NO" having 300 instances. how i can know the dataset is unbalanced and what techniques could be apply using weka to resolve such issue?
Your data set is imbalanced as your class is not a 50/50 or 60/40 distribution. If you use decision trees you might not need to balance your data set. Otherwise you can use under sampling (i.e. use all the smaller class and randomly select same number of majority class several times to make multiple data sets and then combine all classification results) to balance it this is best way for this without loosing information. If you use boosting you could alter the weights and balance data that way. As mentioned above the best way to decide is to explore the data fully. If any other attributes are also imbalanced over their values this will also affect classification results.
If you can to plot those instances using a 2D plot, and if you are able to classify them into two classes just by visual perception, then its quite fine even if the the number instances of two classes are not well balanced. If the instances are scattered enough that it would be difficult to use any non linear classifier, then you have to collect more data samples or use any clustering algorithm to find the cluster centers of two classes.
Use Weka explorer to explore your data. As Bharath said, to balance the classes you will need to add more samples (or remove, not recommended unless this is a toy dataset). Weka explored will show you the 2D plotting of the class distribution.
see the used method in this work :
https://www.researchgate.net/publication/294872631_Une_nouvelle_technique_d%27amelioration_des_mesures_dielectriques_utilisent_la_carte_statistique_du_controle_Application_sur_Ba1-xSrxTiO3
Article Une nouvelle technique d'amélioration des mesures diélectriq...
Your data set is imbalanced as your class is not a 50/50 or 60/40 distribution. If you use decision trees you might not need to balance your data set. Otherwise you can use under sampling (i.e. use all the smaller class and randomly select same number of majority class several times to make multiple data sets and then combine all classification results) to balance it this is best way for this without loosing information. If you use boosting you could alter the weights and balance data that way. As mentioned above the best way to decide is to explore the data fully. If any other attributes are also imbalanced over their values this will also affect classification results.
Thanks Susan Lomax.
kindly tell me if i am using a dataset which have to produced binary class attributes like Student Result "Pass" or :Fail". i have data which have 80% pass students and 20% fail. in reality it is true because same ratio is observed in real life however due to problem of intention the classifier towards majority class here i think needs to be it balanced. the question is that in this case 50/ 50 balancing will be consider right whereas it is impossible or 60/40 should be right or other?
As you will be creating a balanced dataset I would do a 50 50 distribution inthis case and then you could test by classifying the entire dataset through the best model achieved after the usual testing on the 50 50 balanced file. If you create multiple balanced files as described earlier, average out your results and choose your best model. Key is to experiment to see what happens. Usually decision trees do not have problems with a 80 20 distribution but other methods might.
- Badges
- Science topic
- Similar topics
- Computer Science
- Data Mining