Generalized Linear Models (GLM) in Oracle Database Mining (ODM) ostensibly includes the linear model classes, which form a set of restrictive assumptions, and more importantly, classes in which the target (with ‘y’ dependable variable) is generally distributed depending upon the predictor values along with a constant variance, irrespective of that predicted value of the response in question.
The most striking advantage of the linear models and the restrictions thereof is that they inspire computational simplicity and offer an interpretable form of model, along with the option of computing and collating certain diagnostic information regarding the worth or class of the fit.
Now when it comes to GLMs, they let down all these restrictions, thereby making the entire process of ODM less cumbersome and seamless. Furthermore, the sum of all the terms included in a typical linear model can include a set of wide ranges, covering an extremely positive as well as an extremely negative set of values.
Now GLM would accommodate the responses that will under normal circumstances violate the customary assumptions of linear models, through 2 key mechanisms – a Link Function and a Variance Function. While the Link Function would transform the range of target to potentially negative infinity to potentially positive infinity, thereby maintaining the simplest and the most elementary forms of linear models, the Variance Function understandably would express the variance as a typical function of a predicted response, thus accommodating these responses, with a set of non-constant variances, such as binary responses.
Table of Contents
Generalized Linear Models In ODM:
So GLM basically is a parametric technique of modeling, wherein the parametric models make certain assumptions regarding data distribution. When these assumptions are satisfied or met, the parametric models get more efficient than the non-parametric models.
Now the challenge that lies in developing these types of parametric models include a comprehensive assessment of the extent of the assumptions that are met. That is the reason why quality diagnostics are so crucial when it comes to development of quality parametric models, more so in Enterprise Performance Management (EPM), which mainly involves considering the visibility of the operations in a typical closed-loop model, across various aspects of an enterprise.
Transparency And Interpretability:
Models of Oracle Data Mining GLM are pretty simple to interpret. Each and every model generates a wide number of statistics as well as diagnostics. Transparency is another key feature of these models, and the details describe two aspects – the model details describe the key coefficient characteristics and global details come up with high-level statistics.
GLM on the other hand has the ability to predict various confidence bounds. Apart from predicting the best estimates along with a classification probability for each and every row, it also identifies an interval wherein the regression prediction as well as the classification probability will co-exist. The interval width is heavily dependent on the precision of the model and the confidence level that is specified by the end user.
Now, the confidence level in turn can be defined as a measure that calculates the probability of the true value lying within the confidence interval that is computed by the particular model.
These are the models in which the predictors associate with a target, while there is hardly any connection between the predictors themselves. Multicollinearity is another term that is used for describing the multivariate regression with the correlated predictors.
Ridge regression can also be defined as a technique that balances multicollinearity. ODM supports ridge regression for regression as well as classification of the mining functions. The algorithm uses ridge automatically if and when it detects singularity, or in other words exact multicollinearity of data. Information regarding singularity on the other hand is returned in the details of the global model.