You can find all the correctly and incorrectly classified images percentatge per algorithm here.
You can also find all the images that I have used to do this project and additional information like all the algorithm resultss, percentatge of correctly classified images, etc. here
Here you will find the results of the image classification of each algorithm. I have tried the next ones:
BayesNet
It's a bayes algorithm. It provides datastructures and facilities common to bayes Network Learning. It's correctly classified images rate is lower than the incorrectly but has a nice True Positive rate in group 9 with a 0,765. This algorithm is not useful for our project because of it's lower rate of certainly correct classifications.
NaiveBayes
NaiveBayes it's also a bayes algorithm. I'ts rate of correctly classified images it's very similar to de BayesNet but tis is based on the training data. It has a high True Positive rate on the group 4 0,793 so it can be useful to classifie this type of objects (flute).Nevertheless the precision rate are awful so it is not useful because neither classifies correctly separate images.
SimpleLogistic
This one is a 'function' algorithm which has simple regression functions for base learners which is used for fitting the logistic models. It's percentatge of correct classified images is higher than the bayes ones but still being lower than a 50% which is not useful for our classifications. Nevertheless it is very useful to classify 'flute' images with a 0,829 rate (group 4) of True Positive and 'saxophone' images with a 0,704 rate of True Positive classified images (group 9).
IBk
IBk is a 'lazy' algorithm which can select appropiate value of a constant based on cross-validation, can also do distance weighting. It is our first algorithm with a higher than 50% percentatge of correctly classified images with a 51,76% and also has high True Positive rate on the group 9 (saxophone)
KStar
KStar it's also a 'lazy' algorithm but differs from the other one because KStar use an instance-based learner in that it uses an entropy-based distance funcition. We can see that it has a higher percentatge of correctly classified images 53,84%. True Positive rate on both groups nine and four are the higher ones but also we can see that in group 6 (acoustic-guitar) this algorithm has a high precission rate with a 0,841 .
J48
Now, we are entering in 'trees' algorithms. First of all we have J48 which has a lower percentatge of correctly classified images than the other two tree algorithms that we will see later with a 46,67%. Again we can see that it classifies nicely group 9 with a higher than 0,700 True Positive rate and precision rate.
RandomTree
RandomTree is an algorithm used for constructing a tree than considers a constant randomly chosen attributes at each node. For now it has the worst percentatge of correct image classification with a 31,16%. It can be due to it internal form of working which makes a easy relation between images than can be similar but they are actually diffrent things.
RandomForest
RandomForest is based on constructing a forest with random trees it takes a specific feature of an image and when it sees something similar in other one, automatically makes a relation. Apparently it is the better option to classify our images in general because this algorithm has de higher percentatge of correctly classified images with a 58,02% and also has a higher rate of True Positive and precision of an image classification with a 0,802 rate in group 9 (saxophone). This is the most useful algorithm of the chosen ten.
RandomizableFilteredClassifier
Now we eneter in 'meta' algorithms. This one is a variant from FilteredClassifier which instantiates the model with a randomizable filter. It's percentatge of correctly classified imatges its lower than 50% and neither highlits in any group of images or percentatge of precision.
IterativeClassifierOptimizer
This one it's the latest algorithm which optimizes the number of iterations of the given iterative classifier using cross-validation. used and it has a surprising 53,09% of correctly classified imatges and like we have being seeing, the higher True Positive rates are in the group 4 with 0,732 and group 9 with 0,765.
