Clinical Classification
Usage
classification(
data,
dep,
indep,
testSize = 0.33,
noOfFolds = 10,
testing,
reporting = list("classifMetrices"),
classifier,
minSplit = 20,
complexity = 0.01,
maxCompete = 4,
maxSurrogate = 5,
maxDepth = 30,
noOfTrees = 10,
maxDepthRandFor = 30,
sampleFraction = 1,
splitRule,
knnNeighbors = 5,
knnDistance = "euclidean",
svmKernel = "radial",
svmCost = 1,
svmGamma = 1,
plotDecisionTree = FALSE,
predictedFreq = FALSE,
printRandForest = FALSE,
predictedFreqRF = FALSE,
balancingMethod = "none",
validateMethod = "holdout",
bootstrapSamples = 1000,
reportClinicalMetrics = FALSE,
reportConfidenceIntervals = FALSE,
reportMCC = FALSE,
positiveClass = "",
seed = 42,
thresholdMethod = "youden",
thresholdValue = 0.5,
showSummary = FALSE,
showAbout = FALSE,
showGlossary = FALSE
)Arguments
- data
.
- dep
.
- indep
.
- testSize
.
- noOfFolds
.
- testing
.
- reporting
.
- classifier
.
- minSplit
.
- complexity
.
- maxCompete
.
- maxSurrogate
.
- maxDepth
.
- noOfTrees
.
- maxDepthRandFor
.
- sampleFraction
.
- splitRule
.
- knnNeighbors
Number of nearest neighbors for KNN classification.
- knnDistance
.
- svmKernel
.
- svmCost
Regularization parameter for SVM.
- svmGamma
Kernel coefficient for SVM (used in RBF, polynomial, sigmoid kernels).
- plotDecisionTree
.
- predictedFreq
.
- printRandForest
.
- predictedFreqRF
.
- balancingMethod
Method for handling class imbalance in medical datasets.
- validateMethod
Validation method for clinical model assessment.
- bootstrapSamples
Number of bootstrap samples for confidence intervals.
- reportClinicalMetrics
Report sensitivity, specificity, PPV, NPV, and likelihood ratios.
- reportConfidenceIntervals
Include 95\ clinical metrics.
- reportMCC
Calculate Matthews Correlation Coefficient (MCC), a balanced metric for imbalanced datasets. MCC ranges from -1 (perfect disagreement) to +1 (perfect agreement), with 0 indicating random prediction.
- positiveClass
Specify which class should be considered "positive" for clinical metrics (sensitivity, specificity, PPV, NPV, likelihood ratios). If left empty, the second factor level will be used as positive. This is critical for ensuring correct interpretation of diagnostic performance metrics.
- seed
Set random seed for reproducibility. Using the same seed will produce identical results across runs, which is critical for validating analyses.
- thresholdMethod
Method for determining classification threshold for binary outcomes. Youden J maximizes sensitivity + specificity. Manual allows custom cutoff.
- thresholdValue
Custom probability threshold when using manual threshold method. Only used when Decision Threshold Method is set to Manual Cutoff.
- showSummary
Display a plain-language summary of classification results suitable for copying to clinical reports.
- showAbout
Display information about classification methodology and interpretation guidance.
- showGlossary
Display glossary of machine learning and clinical metrics terms.
Value
A results object containing:
results$warnings | a html | ||||
results$modelSettings | a html | ||||
results$confusion$matrix | a table | ||||
results$classificationMetrics$general | a table | ||||
results$classificationMetrics$clinicalMetrics | a table | ||||
results$classificationMetrics$mccTable | a table | ||||
results$classificationMetrics$class | a table | ||||
results$rocCurvePlot | an image | ||||
results$decisionTreeModel | an image | ||||
results$predictedFreqPlot | an image | ||||
results$printRandForest$randomForestModel | a table | ||||
results$text | a html | ||||
results$naturalSummary | a html | ||||
results$aboutAnalysis | a html | ||||
results$glossaryPanel | a html |