Rare Disease Diagnosis with SVM and Specific Target Event

Business Context

A medical research facility is testing a Support Vector Machine (SVM) classifier for a rare disease. They need to validate the model using a low number of folds due to small sample size and must specify the exact target event ('Positive') to ensure correct sensitivity analysis, while keeping logs minimal.

Data Preparation

Creation of a small, specific medical dataset with a text-based target variable.

Copied!

1
2	DATA casuser.rare_disease;
3	INPUT patient_id biomarker_a biomarker_b diagnosis $;
4	DATALINES;
5	1 0.5 1.2 Negative 2 0.8 1.1 Positive 3 0.2 0.9 Negative 4 0.9 1.5 Positive 5 0.4 1.0 Negative 6 0.6 1.1 Negative 7 0.9 1.4 Positive 8 0.3 0.8 Negative ;
6
7	RUN;
8

Étapes de réalisation

Ensure data is loaded and promoted for access.

Copied!

1
2	PROC CAS;
3
4	TABLE.promote name="rare_disease" caslib="casuser";
5
6	QUIT;
7

Execute crossValidate with 'SVM', minimal logging (logLevel=0), specific target event, and kFolds=2 (minimum allowed).

Copied!

1	PROC CAS;
2	mlTools.crossValidate /
3	TABLE={name="rare_disease"}
4	modelType="SVM"
5	kFolds=2
6	logLevel=0
7	targetEvent="Positive"
8	casOut={name="cv_svm_diagnosis", replace=TRUE}
9	trainOptions={
10	target="diagnosis",
11	inputs={"biomarker_a", "biomarker_b"},
12	nominals={"diagnosis"}
13	};
14	QUIT;

Expected Result

The action executes silently (no detailed logs) due to logLevel=0. It successfully performs a 2-fold cross-validation using the SVM algorithm. The 'Positive' value is correctly used as the event of interest for calculating statistics like misclassification rate.

Voir la documentation technique de crossValidate