BSNS112: Exam - Nonlinear and
computational modelling approaches
Questions and Answers.
input variables (x) - Answer explanatory/ independent
output variables (y) - Answer response (dependant)
poor data quality - Answer missing values and labels, incorrect data entry, mixing of
numerical and text recordings for variables and inconsistent values
high quality data - Answer no missing values and labels, data correctly recorded, no mixed
types of recordings and has been checked for error (validated)
Occam's Razor - Answer the simpler of two explanations for an occurrence that perform
similarly is preferred
dummy variables - Answer a numeric variable that represents categorical data, such as
gender, race, political affiliation, etc.
Backward Stepwise Regression - Answer Start with all features (explanatory variables)
1. Build the regression model
2. remove the worst variable (i.e. the one with a p-value > 0.05)
3. Repeat from step 1 until all variables are significant at some level (probably < 0.05)
Forward Stepwise Regression - Answer Start with an empty model
1. add the 'best' variable each time until no improvement or significant variable
Logistic Regression - Answer using a (binomial) logistic (link) function with a generalised
linear model
- a logistic function gives outputs for 0 and 1 for all values of explanatory variables
- therefore can be included to model the probability of categorical responses occurring
Automatically Transformed Linear Model (ATLM) - Answer created as a baseline model
requiring no human interaction that would perform reasonably well for comparison against
Software Effort Estimation (SEE)
- based on the idea that MLR's work better when the variables are normally distributed
computational modelling approaches
Questions and Answers.
input variables (x) - Answer explanatory/ independent
output variables (y) - Answer response (dependant)
poor data quality - Answer missing values and labels, incorrect data entry, mixing of
numerical and text recordings for variables and inconsistent values
high quality data - Answer no missing values and labels, data correctly recorded, no mixed
types of recordings and has been checked for error (validated)
Occam's Razor - Answer the simpler of two explanations for an occurrence that perform
similarly is preferred
dummy variables - Answer a numeric variable that represents categorical data, such as
gender, race, political affiliation, etc.
Backward Stepwise Regression - Answer Start with all features (explanatory variables)
1. Build the regression model
2. remove the worst variable (i.e. the one with a p-value > 0.05)
3. Repeat from step 1 until all variables are significant at some level (probably < 0.05)
Forward Stepwise Regression - Answer Start with an empty model
1. add the 'best' variable each time until no improvement or significant variable
Logistic Regression - Answer using a (binomial) logistic (link) function with a generalised
linear model
- a logistic function gives outputs for 0 and 1 for all values of explanatory variables
- therefore can be included to model the probability of categorical responses occurring
Automatically Transformed Linear Model (ATLM) - Answer created as a baseline model
requiring no human interaction that would perform reasonably well for comparison against
Software Effort Estimation (SEE)
- based on the idea that MLR's work better when the variables are normally distributed
