SOA EXAM PA QUESTIONS AND
ANSWERS
Random Sampling - Correct Answers -all members of the population have an equal
chance of being selected
Stratified Sampling - Correct Answers -A type of probability sampling in which the
population is divided into groups with a common attribute and a random sample is
chosen within each group
Systematic Sampling - Correct Answers -Observations are drawn according to a set
pattern and there is no random mechanism controlling which observations are sampled
Model Selection using AIC - Correct Answers -All available data is used to train the
potential models. This method calculates how well it will fit the training data, with a
preference for models with fewer predictors. Because this method does not directly
consider how well the models fit new data, it does not give the user direct insight into
how well the model generalizes to unseen data.
Structured Data - Correct Answers -Data that (1) are typically numeric or categorical; (2)
can be organized and formatted in a way that is easy for computers to read, organize,
and understand; and (3) can be inserted into a database in a seamless fashion.
Unstructured data - Correct Answers -Nonnumeric information that is typically formatted
in a way that is meant for human eyes and not easily understood by computers
Univariate data exploration in R - Correct Answers -Numeric Variables: Use summary()
function or histogram/boxplots
Categorical Variables: Use table() function or bar charts (could use normal frequency
counts or proportions)
What to do with a right-skewed target variable - Correct Answers -Apply a monotone
concave function to shrink the outlier values and symmetrize the overall distribution
while preserving the ranks of the observed values of the variable. Transformations
include:
- Log transformation - can be applied only if the variable of interest is strictly positive
(not = 0!)
- Square-root transformation - can be applied to non-negative variables (can equal 0)
, In general, the log transformation does a better job of remedying the right skewness of a
variable, but it may overdo things and make the transformed variable left-skewed.
Bivariate data exploration - Correct Answers -Numeric vs Numeric: Visualize in
scatterplot, look at correlation table (use cor() function)
Numeric vs Categorical: Visualize with split box plots
Categorical vs Categorical: Two way frequency table (use table() function with 2
variables) or split bar charts
Target variable - Correct Answers -the variable we are interested in predicting
Explanatory variables - Correct Answers -the variables used to predict the target
variable, also known as predictors or features
Supervised learning problems - Correct Answers -there is a target variable guiding our
analysis, and our goal is to understand the relationship between the target variable and
the predictors, and/or make accurate predictions for the target based on the predictors.
GLMs and decision trees fit into this category
Unsupervised learning problems - Correct Answers -a target variable is absent and we
are interested in extracting relationships and structures between different variables in
the data. Principal components analysis and cluster analysis fit into this category
Model Selection using a training/test split - Correct Answers -Use a large percentage of
the data to train the potential models and a smaller percentage to test the models.
Because the test data was not used to train the models, it can be used to directly
assess how well each model makes predictions when presented with new data. The
testing is only done once, though, and some data will never be used to train the
model/some will never be used to train the model.
Root mean squared error (RMSE) - Correct Answers -A common numeric predictive
metric that aggregates all of the prediction errors on the test set and provides an overall
measure of predictive accuracy. The smaller the value, the more predictive the model.
The metric has the same unit as the target variable, which makes it easy to interpret as
the absolute value of a typical prediction error.
Misclassification error rate - Correct Answers -A common categorical variable predictive
metric that calculates the proportion of misclassified observations on the test set. The
smaller the rate, the more predictive the classifier.
Model Selection using n-fold Cross validation - Correct Answers -Splits the dataset into
a training and test group to train the model and later test the model. The split is done n
times on the same data such that all the data is used at some point for both training and
test models. While this method takes longer to carry out, it reduces overfitting.
ANSWERS
Random Sampling - Correct Answers -all members of the population have an equal
chance of being selected
Stratified Sampling - Correct Answers -A type of probability sampling in which the
population is divided into groups with a common attribute and a random sample is
chosen within each group
Systematic Sampling - Correct Answers -Observations are drawn according to a set
pattern and there is no random mechanism controlling which observations are sampled
Model Selection using AIC - Correct Answers -All available data is used to train the
potential models. This method calculates how well it will fit the training data, with a
preference for models with fewer predictors. Because this method does not directly
consider how well the models fit new data, it does not give the user direct insight into
how well the model generalizes to unseen data.
Structured Data - Correct Answers -Data that (1) are typically numeric or categorical; (2)
can be organized and formatted in a way that is easy for computers to read, organize,
and understand; and (3) can be inserted into a database in a seamless fashion.
Unstructured data - Correct Answers -Nonnumeric information that is typically formatted
in a way that is meant for human eyes and not easily understood by computers
Univariate data exploration in R - Correct Answers -Numeric Variables: Use summary()
function or histogram/boxplots
Categorical Variables: Use table() function or bar charts (could use normal frequency
counts or proportions)
What to do with a right-skewed target variable - Correct Answers -Apply a monotone
concave function to shrink the outlier values and symmetrize the overall distribution
while preserving the ranks of the observed values of the variable. Transformations
include:
- Log transformation - can be applied only if the variable of interest is strictly positive
(not = 0!)
- Square-root transformation - can be applied to non-negative variables (can equal 0)
, In general, the log transformation does a better job of remedying the right skewness of a
variable, but it may overdo things and make the transformed variable left-skewed.
Bivariate data exploration - Correct Answers -Numeric vs Numeric: Visualize in
scatterplot, look at correlation table (use cor() function)
Numeric vs Categorical: Visualize with split box plots
Categorical vs Categorical: Two way frequency table (use table() function with 2
variables) or split bar charts
Target variable - Correct Answers -the variable we are interested in predicting
Explanatory variables - Correct Answers -the variables used to predict the target
variable, also known as predictors or features
Supervised learning problems - Correct Answers -there is a target variable guiding our
analysis, and our goal is to understand the relationship between the target variable and
the predictors, and/or make accurate predictions for the target based on the predictors.
GLMs and decision trees fit into this category
Unsupervised learning problems - Correct Answers -a target variable is absent and we
are interested in extracting relationships and structures between different variables in
the data. Principal components analysis and cluster analysis fit into this category
Model Selection using a training/test split - Correct Answers -Use a large percentage of
the data to train the potential models and a smaller percentage to test the models.
Because the test data was not used to train the models, it can be used to directly
assess how well each model makes predictions when presented with new data. The
testing is only done once, though, and some data will never be used to train the
model/some will never be used to train the model.
Root mean squared error (RMSE) - Correct Answers -A common numeric predictive
metric that aggregates all of the prediction errors on the test set and provides an overall
measure of predictive accuracy. The smaller the value, the more predictive the model.
The metric has the same unit as the target variable, which makes it easy to interpret as
the absolute value of a typical prediction error.
Misclassification error rate - Correct Answers -A common categorical variable predictive
metric that calculates the proportion of misclassified observations on the test set. The
smaller the rate, the more predictive the classifier.
Model Selection using n-fold Cross validation - Correct Answers -Splits the dataset into
a training and test group to train the model and later test the model. The split is done n
times on the same data such that all the data is used at some point for both training and
test models. While this method takes longer to carry out, it reduces overfitting.
