ISYE 6501 Midterm 1 Exam Questions And Answers
Verified 100% Correct
Change detection - ANSWER Identifying when a significant change has taken place in
a process.
Classification - ANSWER The separation of data into two or more categories, or (a
point's classification) the category a data point is put into.
Classification tree - ANSWER Tree-based method for classification. After branching to
split the data, each subset is analyzed with its own classification model.
Classifier - ANSWER A boundary that separates the data into two or more categories.
Also (more generally) an algorithm that performs classification.
Clustering - ANSWER Separation of data points into groups ("clusters") based on 5
nearness/similarity to each other. A common form of unsupervised learning
Confusion matrix - ANSWER Visualization of classification model performance
Cross-validation - ANSWER Validation technique where a model is tested on data
different from what it was trained on
CUSUM - ANSWER Change detection method that compares observed distribution
mean with a threshold level of change. Short for "cumulative sum".
Regression can ANSWER two questions - ANSWER - How systems
work
- What will happen in the future
Regression - ANSWER Statistical model that describes relationships between
variables, and/or predicts future values of a response..
Simple Linear Regression - ANSWER - Linear regression with one predictor
- Look for Linear relationship between predictor and response
Linear Regression - ANSWER Regression model where the relationships between
attributes and a response are modeled as linear functions: 𝑦𝑦 = 𝑎𝑎0 + ∑ 𝑎𝑎𝑖𝑖𝑥𝑥𝑖𝑖 𝑚𝑚
𝑖𝑖=1 .
, likelihood - ANSWER - measure the probability (density) for any parameter set
maximum liklihood - ANSWER parameters that give the highest probability - A
method that finds the set of parameter values for which a model is most likely to
generate the actual values of the data
maximum likelihood fitting - ANSWER Regression with independent normally
distributed errors
Can get complex fast
Different estimation formulas
Different assumptions about the error
Good statistical software
Can handle more complexity than simple regression
Akaike Information Criterion (AIC) - ANSWER L* is maximum likelihood value
K is number of parameters being estimated
AIC = 2k-2ln(L*)
Penalty term: balances likelihood with simplicity
Helps avoid overfitting
Prefer models with smaller AIC
Nice properties if there are infinitely many data points
Corrected AIC - ANSWER Improved version of AIC, especially when sample size is
small
P Values - ANSWER Estimated the probability: the coefficient = 0
Hypothesis testing
P-value > .05
Remove the corresponding attribute from the model
Other thresholds besides .05 can also be used
Higher threshold: more factors can be included: possibility of including irrelevant factor
Lower threshold: less factors can be included: possibility of leaving out a relevant factor
Two warnings
With large amounts of data, p values get small even when attributes are not at all
related to the response
P values are only probabilities even when meaningful
p-value - ANSWER (1) In hypothesis testing, probability that results at least as extreme
as those in the data would be observed if the null hypothesis is true. (2) In regression,
probability that results at least as extreme as those in the data would be observed if the
coefficient of a variable is zero.
Verified 100% Correct
Change detection - ANSWER Identifying when a significant change has taken place in
a process.
Classification - ANSWER The separation of data into two or more categories, or (a
point's classification) the category a data point is put into.
Classification tree - ANSWER Tree-based method for classification. After branching to
split the data, each subset is analyzed with its own classification model.
Classifier - ANSWER A boundary that separates the data into two or more categories.
Also (more generally) an algorithm that performs classification.
Clustering - ANSWER Separation of data points into groups ("clusters") based on 5
nearness/similarity to each other. A common form of unsupervised learning
Confusion matrix - ANSWER Visualization of classification model performance
Cross-validation - ANSWER Validation technique where a model is tested on data
different from what it was trained on
CUSUM - ANSWER Change detection method that compares observed distribution
mean with a threshold level of change. Short for "cumulative sum".
Regression can ANSWER two questions - ANSWER - How systems
work
- What will happen in the future
Regression - ANSWER Statistical model that describes relationships between
variables, and/or predicts future values of a response..
Simple Linear Regression - ANSWER - Linear regression with one predictor
- Look for Linear relationship between predictor and response
Linear Regression - ANSWER Regression model where the relationships between
attributes and a response are modeled as linear functions: 𝑦𝑦 = 𝑎𝑎0 + ∑ 𝑎𝑎𝑖𝑖𝑥𝑥𝑖𝑖 𝑚𝑚
𝑖𝑖=1 .
, likelihood - ANSWER - measure the probability (density) for any parameter set
maximum liklihood - ANSWER parameters that give the highest probability - A
method that finds the set of parameter values for which a model is most likely to
generate the actual values of the data
maximum likelihood fitting - ANSWER Regression with independent normally
distributed errors
Can get complex fast
Different estimation formulas
Different assumptions about the error
Good statistical software
Can handle more complexity than simple regression
Akaike Information Criterion (AIC) - ANSWER L* is maximum likelihood value
K is number of parameters being estimated
AIC = 2k-2ln(L*)
Penalty term: balances likelihood with simplicity
Helps avoid overfitting
Prefer models with smaller AIC
Nice properties if there are infinitely many data points
Corrected AIC - ANSWER Improved version of AIC, especially when sample size is
small
P Values - ANSWER Estimated the probability: the coefficient = 0
Hypothesis testing
P-value > .05
Remove the corresponding attribute from the model
Other thresholds besides .05 can also be used
Higher threshold: more factors can be included: possibility of including irrelevant factor
Lower threshold: less factors can be included: possibility of leaving out a relevant factor
Two warnings
With large amounts of data, p values get small even when attributes are not at all
related to the response
P values are only probabilities even when meaningful
p-value - ANSWER (1) In hypothesis testing, probability that results at least as extreme
as those in the data would be observed if the null hypothesis is true. (2) In regression,
probability that results at least as extreme as those in the data would be observed if the
coefficient of a variable is zero.
