I am trying to classify a data using Supervised machine learning algorithms.
Everything's working fine, but just for my curiosity, I tried 6 classification algorithms simultaneously on a single data set. Steps followed are as follows-
1> Train all the algorithms.
2> predicted the result(either 1 or 0) for all test_data individually, by all algorithms.
3> If most of the algos gave 0, i considered the result for that data pair to be 0, similarly for result 1.
4> Then i found out the overall accuracy.
I expected the overall accuracy to be higher then the individual results(By each algorithm working individually), But i got almost the average accuracy.(Average here means average of accuracies of individual algos).
Can anyone please help me to find the reason?
This depends on the algorithms you picked. Many algorithms are sensitive to different things. For instance, k-means, linear SVM, and power iteration clustering will get markedly different results.
You got what you asked for: you averaged the votes, without coordinating the algorithms in any way. You got an average result.
I doubt that weighted averaging will help much; all you're doing there is training a meta-model. Instead, consider the data set you have. You need to research modelling algorithms and pick one that tends to work well on the statistical shape of your data set with respect to the desired purpose. Since you've given us none of this background, we can't help with specifics.