OMSA Midterm Exam Questions With Complete Solutions Latest Updated 2024 (Graded )

OMSA Midterm Exam Questions With Complete Solutions Latest Updated 2024 (Graded ) Overfitting - Answer- Number of factors is too close to or larger than number of data points -- fitting to both real effects and random effects. Comes from including too many variables! Ways to avoid overfitting - Answer- - Need number of factors to be same order of magnitude as the number of points - Need enough factors to get good fit from real effects and random effects Simplicity - Answer- Simple models are better than complex. When fewer factors exist, less data collection is required -- less chance for including factors that are not significant. Another reason for variable selection! DOE - Answer- systematic method to determine the relationship between factors affecting a process and the output of that process. must make sure either: 1) 2 data sets have same mix 2) break down data into smaller tests that test all factors, not just one. forward selection - Answer- go step by step either narrowing or building a model -- begin with no factors. only allow new factors with p-value 0.1 or lower and removing any factors above 0.05. Greedy models - Answer- forward selection, backward selection, stepwise regression means: at each step it does the one thing that looks best withouth taking future options into consideration scaled variable selection models - Answer- lasso, elastic net, (and ridge even though not variable selection) backward selection - Answer- start with model includes all factors and at each step find worst factor and remove it from the model. continue till there's no factor bad enough to remove and model doesn't have any more factors we want. step-wise regression - Answer- combination of forward and backward elimination -- beginning with either all factors or no factors. after adding each new factor (and at end) -- eliminate right away any factors that no longer appear any good. future options are not considered. could use AIC or BIC, or model's R squared to pick factor to add/remove in fwd, bckwd, or stepwise lasso method - Answer- adds constraint tau to std reg model. chooses coefficients to min sum of squared errors. right value of tau dependent on 2 things: 1) # of variables you want 2) quality of the model as you allow more variables Correlation -- picks just one to have non-0 coefficient. other is left out. elastic net - Answer- constrain combination of abs value and coefficients and their squares -- mix of ridge and lasso. need to scale and pick right value of tau + lambda ridge regression - Answer- similar to elastic net, but without abs value term (lasso portion). adds quadratic term that shrinks coefficient values (pushing reg coeff to 0 or regularizes them). this adds bias but shrinks variance as cannot shrink to 0 as easily as quadratic term trades off bias for variance but can lead to better predictive models. blocking - Answer- arranging of experimental units in groups that are similar to one another. accounts for bias differences via variable differences that are dependent on dependent variable accounts for differences can reduce variability in our estimates