Neural Networks with a Principal Component Step
Run PCA on a dataset, then use it in a neural network model
## S3 method for class 'default': pcaNNet(x, y, thresh = 0.99, ...) ## S3 method for class 'formula': pcaNNet(formula, data, weights, ..., thresh = .99, subset, na.action, contrasts = NULL)
## S3 method for class 'pcaNNet': predict(object, newdata, type = c("raw", "class"), ...)
- A formula of the form
class ~ x1 + x2 + ...
- matrix or data frame of
xvalues for examples.
- matrix or data frame of target values for examples.
- (case) weights for each example -- if missing defaults to 1.
- a threshold for the cumulative proportion of variance to capture from the PCA analysis. For example, to retain enough PCA components to capture 95 percent of variation, set
thresh = .95
- Data frame from which variables specified in
formulaare preferentially to be taken.
- An index vector specifying the cases to be used in the training sample. (NOTE: If given, this argument must be named.)
- A function to specify the action to be taken if
NAs are found. The default action is for the procedure to fail. An alternative is na.omit, which leads to rejection of cases with missing values on any required variable. (NOTE: If given, this
- a list of contrasts to be used for some or all of the factors appearing as variables in the model formula.
- an object of class
nnetas returned by
- matrix or data frame of test examples. A vector is considered to be a row vector comprising a single case.
- Type of output
- arguments passed to
The function first will run principal component analysis on the data. The cumulative percentage of variance is computed for each principal component. The function uses the
thresh argument to determine how many components must be retained to capture this amount of variance in the predictors.
The principal components are then used in a neural network model.
When predicting samples, the new data are similarly transformed using the information from the PCA analysis on the training data and then predicted. Because the variance of each predictor is used in the PCA analysis, the code does a quick check to make sure that each predictor has at least two distinct values. If a predictor has one unique value, it is removed prior to the analysis.
Ripley, B. D. (1996) Pattern Recognition and Neural Networks. Cambridge.
data(BloodBrain) modelFit <- pcaNNet(bbbDescr, logBBB, size = 5, linout = TRUE, trace = FALSE) modelFit predict(modelFit, bbbDescr)