KNEAREST NEIGHBOUR Sample Clauses
KNEAREST NEIGHBOUR. KNearest neighbour (kNN) is a classification decision system, which makes use of distance measures within an ndimensional space. The system finds matches for new points within the search space by consulting neighbouring points, assessing the neighbouring point’s class and assigning the new point a class based on this. "K", in the case of kNN refers to the number of points to be consulted before a classification is determined. The equation for finding a basic kNN based classification is as follows:
KNEAREST NEIGHBOUR. This technique formed the main stay of the distance classification system used whenever the representation involved using a vector space. This technique is widely used by many different researchers to great affect, in many different circumstances. In this instance, the value of “k” was not resolved through testing. This naturally presents a significant problem when proposing a bestcase classification system. The three options available based on the results obtained are to set k equal to one, making this a nearest neighbour classification system. This option negates the use of a kNN classification system. The second option is to set k to an arbitrary value between one and ten, though probably towards the higher end of the spectrum, as the results do seem to suggest an increase in classification rates at this end. This proposal would not necessarily resolve the best classification rates, and would rely on the position of the underlying data. The third option would be to use a weighted knearest neighbour system and simply use all the available reference points within the vector space. Although this may increase classification cost marginally, the problem of the underlying data structures would be effectively removed. Although the testing data given does not conclusively determine the correct value of k, this classification system will almost certainly be used within this project, as it provides an accurate, fair and computationally cheap system for discerning the class of new objects based on known training points, especially when weighted kNN is used.