Question

3. Suppose you were running the Adaboost algorithm for a two-class classification problem, but one of the data points is "difficult to classify" because it is far from the other points in its class and lies inside a cluster of points belonging to the opposite class. Your friend argues that the data weighting coefficient for this point will decrease because the AdaBoost algorithm will try to discount this point as an outlier. Do you agree or disagree and why?

Answer 1

Answer:

Here Adaboost or "Adaptive algorithm " is the principal functional boosting calculation.

It has order issues and expects to change over a lot of feeble classifiers into a solid one.

Boosting is method which centers around making a solid classifier from various powerless classifiers.

AdaBoost is the primary effective boosting calculation created for double order.

Each case in preparing informational index is weighted. The underlying weight is given as follows

i.e.,

weight(xi) = 1/n

Where as the xi is the i'th preparing case and n is the quantity of preparing occurrences.

The principle thought is to set loads to the two classifiers and information focuses so that it powers classifiers to focus on perceptions that are hard to effectively characterize.

Her this procedure is done successively way with the end goal that two loads are balanced at each progression as cycles of the calculation continues.

Loads of the accurately ordered focuses are not diminished. Just the loads of the inaccurately characterized focuses are expanded.

The new classifier may order the old focuses inaccurately.

In any case, past adaptations of the classifier are not discarded.

In the event that we have an anomaly that is difficult to group effectively, the exception will aggregate a great deal of weight.

The classifier will be compelled to offer need to the point and order it accurately. In any case, this present classifier's feeling isn't be so significant at last on the grounds that just a single point was characterized accurately (exception).

It is a decent method to recognize exceptions. Simply, discover the focuses with extremely huge loads.

The calculation expresses that the examples in the preparation set are weighted each time they are effectively arranged or misclassified to guarantee that consequent students center around occasions that are difficult to group.

It is altogether founded on the preparation dataset.

With the above component of AdaBoost you can differ that "that the information weighting coefficient for this point will diminish in light of the fact that the AdaBoost calculation will endeavor to limit this point as an anomaly."