Research - (2020) Volume 13, Issue 4
This work deals with the problem of knowing a group of people that adequately responds to a specific treatment in order to classify community in groups is the objective. In the process to make this classification, a lot of work is necessary to analyze the results from the cluster analysis and obtain the minimal parameters that define a specific group that can be classified as treatment target. Also is presented the algorithm called Custom-clustering to solve this problem.
Classification • Clustering • COVID-19
Classification of objects into meaningful groups is a central task in Science . Cluster analysis is a statistical technique specialized to classify units into groups. Although cluster analysis is widely employed in other disciplines, it’s used in Political Science, sociology, empirical research, then using factor analysis and other multivariate statistical techniques . The principal aim of this paper is to present an algorithm to cluster analysis for medical treatment or political interest.Difference to neural network in AI, they try to adjust one Hypothesis to a neural net , adjusting weights to the examples, otherwise the clustering method give the true hypothesis from de experimental data. For a case to identify a chair, is necessary to deeply search in the NN to discover the weight for a two foot in a chair is 0, but in clustering there are two different areas 1 and 3 to 4 foot, at the end takes the same results from the same data.
Multiple treatments are being experimented to reduce or cure the effects of COVID-19 in humans, a new method starting from a plasma with highpower immune agents, Grifolsenterprise produce Hyperimmune globulins designed to give a patientimmunity. From which it must be determined accurately the population that can be used with good results. One method to determine the limits of the set of population that is determined by certain characteristics that defines it, is the classification by clustering. In this work we present a classification algorithm using clustering function that in the first part of the algorithm adjusts the parameters to control the group at desired target specification, and the second part determines the minimum necessary characteristics to define it. Unlike the main PCA component analyzes, which determines a linear combination of the axes or parameters that determines the group, the resulting axes do not represent real values of characteristics of the subject, and neither to adjust the cluster to some specification.We will start from possible characteristics that define each subject, and through sampling we obtain a representation of the possible spectrum of people to whom the treatment has been applied, and from there we look for the population group that best accepts the treatment, it is called classification with labels, which can be: negative, in the case of harming the patient, positive otherwise, and neutral without effect on the subject. The objective is to determine apopulation group where the treatment is adequate and necessary, within desired margins age, harm, etc. and with sufficient representation, sampling to be able to consider the results valid and representative.
In many situations the superficial analysis of the results usually leads to say that they are not conclusive, because certain parameters indicate good performance in some range, and scattered in another, this does not mean that this parameter is not useful. It can be possible that said variable depends at the same time on another, so when representing the results in 2 dimensions it can be seen in some cases that the second variable displaces the results and discriminates the positives from the negatives according to their magnitude. Finding these relationships, in a multi-related and multidimensional space, is the objective of this work, and in the end to enclose the possible group among the indispensable parameters.For this, the maximum number of variables that could have information to define the group are collected, such as age, blood group, diseases suffered, and active, treatments, etc. This does not allow a representation by color results to visualize the position of the group, due to the multi-dimensionality of the data. We must analyze the data by the algorithm provided to enclose and reduce the variables to the minimum necessary.
The name of Custom-clustering came from the possibility of the algorithm to close the group of individuals as desired. Difficulty to discover small group into the middle of the others big groups, and also depending on the sample density, is shown in the Figure 1. Is easy to imagine a group into 2 dimension representation and in colored results, but normally the data is multidimensional, with axes with information and others no. The work presented is to analyze these axes and eliminate the non-profitable, and find the group with a validating measure. We use cluster function provided from Matlab .
T=Cluster (Z,' MaxClust', N) constructs a maximum of N clusters using distance as a criterion.
The function permits to find a numbers of groups depending of the N parameter, and modifying it permits to adjust one desired group as fine as possible.
The difficulty to obtain at this time true data for use the algorithm in health problem make us to demonstrate their use into the problem of social interest, useful for commercial and electoral activities as a new paradigm came from the social media, which generates big amount of information from the users according its movements into the smartphones apps.
Some different situations to prevent and avoid the non-stabilized position, looking into the data is possible to advise the robot is approximated to one identified cluster meaning a dangerous situation, and the control must to react to move the robot againstthis cluster, for example, changing velocity or inclination.