BSNS112 - Final Exam Prep Questions With Verified Answers.

BSNS112 - Final Exam Prep Questions With Verified Answers. What does quantitative data use? - answermeans What is discrete data? - answerQuantitative. Measured in specific values What is continuous data? - answerQuantitative. Measure in infinite values What does qualitative data use? - answerProportions What is ordinal data? - answerQualitative. Conveys a ranking What is nominal data? - answerQualitative. Uses labels (no ranking) confidence interval for one proportion - answerq-hat = (1-p-hat) hypothesis test for one proportion - answer Confidence interval for the difference in two proportions - answer sample size for estimating mean - answerE = B (on formula sheet) standard normal transformation formula - answercalculating when Z is unknown (or x or sd but most likely z) right skewed distribution - answermean median, also known as a positive skew left skewed distribution - answermean median, also known as a negative skew 90% confidence interval - answer1.645 95% confidence interval - answer1.96 99% confidence interval - answer2.575 90% confidence interval of the proportion of all market goers who are students would be: - answerNarrower than the 95% confidence interval When do you reject the null hypothesis? - answerWhen the p value is less than 0.05 (p-value low, reject that SHO!!!) 5% level of significance What does the t-value represent? - answerthe sample means is (x) amount standard errors more/less (depending if it is positive or minus) than the hypothesised mean What is a type one error? - answerRejecting the null hypothesis when it is true What is a type two error? - answerfailing to reject a false null hypothesis p-value - answer-is calculated based on the assumption that the null hypothesis is true -the p-value is sample specific, meaning if you collected another random sample of the same size from the same population, the p-value would likely be different. Stratifed random sample - answerFor the same sample size, parameter estimates are usually more accurate than for simple random sampling Categorical variables (qualitative variables) - answerthose that divide subjects into groups, but do not allow any sort of mathematical operations to be performed on the data Numerical Variables (Quantitative) - answernumbers Ordinal - answerrank, order nominal variables - answervariables measured in monetary units....currency Data quality and feature selection - answer input variable - answerxi - explanatory variable (independent) variables these are also called features output variable - answerYk - the response (dependent) variables issues with data - answer- data you tend to play with (number you're given in a sample) - features and examples (more of one less of the other) - large number of example of features - very large number of examples data quality - answer- most often comes as a table - but this isn't the case (in the real world) - data can have holes and look and it won't look clean (complex format) poor data quality - answer-missing column variables -missing values - errors in the data entry - mixed numeric and test - inconsistent values Transformations (scaling) - answer- reduce the scale (range) of the data - transform the mean, 0, and SD, 1 Log transform - answer- some data is highly skewed and is better if you change it to look more normally distributed - square root helps Feature selection reduction - answerworst case, more explanatory variables than examples, so a multiple linear regression cannot be constructed - simple method 1. select the feature F that is most correlated with the response 2. remove all features that are highly correlated w/ F (above some given time hold) 3. repeat steps 1 and 2 until number of features is manageable Correlation - answerCorrelation is NOT causation Cor X,Y/óxóy stepwise multiple regression model - answermulitiple factors the impact the regression (mulitiple dimension model) backward multiple regression model - answer- build regression - price worst variable - take variable out confirmed adjusted r-squared hasn't gone down (quantity and fit of the model) Principal component analysis - answer- very common method for dimensionaity reduction (i.e. reduce the number of features) - rotates the data map more complicated equations though transforming the data - non-linear models - answerstraight line sometimes not best fit - take the x-value and make new deviratives Y-hat = Bo+ B1 X + B2X^2 + B3X^3 Empirical Risk - answer-model has bee measured from the data it was built on - tends to exaggerate the model performance - subject to memorisation (overfitting)