Patient Specific Survival Prediction - Option Details

This page provides details about options available to you for training a PSSP predictor

---

Feature Standardization

Checking this option will standardize each feature in your data set. This will generally speed up training, and sometimes increase accuracy.

---

Cross-Validation

Checking this option will run cross-validation on your data set. This will allow you to examine your training data, and get an estimate of your predictor's accuracy.

---

Tuning Parameters

Checking this option will try to improve the accuracy of your predictor by evaluating various parameter settings. This will increase training time.

---

Feature Imputation

If your data set contains missing values, they will be replaced with the column mean.

---

Regularization

This option specifies the penalty term in the optimization function. Available regularizations are: L1, and L2. L1 regularization will result in a sparse feature set (i.e. a predictor that uses less of the features), but the time to train a predictor is longer.

Smoothed Log-Likelihood

This option uses a smoothed PDF function to calculate the log-likelihood, and also uses this to tune parameters.