ISYE 6501 EXAM QUESTIONS WITH 100% SOLUTIONS 2024

Support Vector Machine(SVM) is a supervised machine learning algorithm used for? - ANSWER Classification How to split the data if we only have one model? - ANSWER 70% training data, 30% testing data How to split the data if we want to compare models? - ANSWER 70% training, 15% validation and 15% testing When do we need to do scaling in data? - ANSWER When our factors/attributes/dimensions are orders of magnitude different such as income vs. credit score (income is much much larger) Which type of classifier tend to be more overfitting? - ANSWER Soft classifier What is a measure of quality used in linear regression? - ANSWER R-squared or adjusted R-squared: shows how well your model explains the observed data. The higher R-squared, the better the model is expected to perform AIC, BIC: The lower AIC and BIC, the better the model fits Select where each data is categorical a. Education level b. GPA c. Gender d. Temperature - ANSWER a and c What type of data is used in CUSUM? - ANSWER time series data What type of data is used in Exponential Smoothing model? - ANSWER time series data What is GARCH do? - ANSWER variance estimation For KNN and K-means clustering, why is scaling important - ANSWER Otherwise some variables could be associated with more 'weight' reference: means-8129e4d88ed7 What are some approaches to check if outliers are successfully removed - ANSWER e.g., histogram, what are some examples of classification methods we've learned that would fall under the supervised vs. unsupervised algorithms - ANSWER e.g., PCA = unsupervised SVM, KNN = supervised other examples? In the SVM equation, what does the mj variable do? specifically, the mj varaible that's in front of the max { } portion of the equation - ANSWER mj 1 for more costly errors, and mj 1 for less costly errors Reference: See ISYE-6501 Module 2 Playlist v0822. Video length 5:49; time-stamp 1:53 in KNN, what's the difference between 1-norm and 2-norm? i.e., L1 vs. L2 - ANSWER Different ways of measuring the distance. 1-norm or L1 is the manhattan difference which is based on the difference in terms of x and y. | x1-x2 | + | y1-y2 | where as the L2 is the euclidean distance which is the sqrt((x1-x2)^2 + (y1-y2)^2) what does it mean if k = 1 in the context of knn - ANSWER it means you'll predict a unknown data point as the same class as its single closest neighbor is k-means algorithm heuristic? yes or no? and if yes, what does it mean? - ANSWER yes. heuristic = "fast, good, but not guaranteed to find absolute best solution" Reference: M4L3 4:22/5:27 time stamp if there are 100 samples and we've decided to do a 5 fold cross-validation, how many training and test samples will be in each fold - ANSWER 5 fold = 5 'parts', so 4/5 part will be used for each training (n=80) and the remaining 1/5 part will be used for testing what is a common % of data split for training vs. validation? what about training vs. validation vs. testing? - ANSWER around 70/30 common for training/validation. 70/15/15 or somewhere around that number is common for training vs validation vs testing if someone describes a time series model as 'double exponential smoothing', which components will be involved? which one will NOT be involved - ANSWER level and trend will be involved. seasonality would not be how are ARIMA and GARCH similar? how are they different? - ANSWER ARIMA and GARCH both can be used for time series data and smoothing. assuming ARIMA(p, d, q), what values for p, d, and q would essentially represent a basic exponential smoothing model - ANSWER ARIMA (0, 1, 1) Reference = lecture slides When does ARIMA perform better than exponential smoothing in terms of short-term forecasting? how many samples are generally required for ARIMA model? - ANSWER when the data is more stable, with fewer peaks, valleys, and outliers. want 40 PAST data points for ARIMA to work well. which set of the following equation is used to detect a decrease vs. increase in the context of CUSUM analysis St = max{0, St-1 + (Xt - u - C)} Is St = T vs St = max(0, St-1 + (u - Xt - C)} Is St = T? - ANSWER Detecting an increase St = max{0, St-1 + (Xt - u - C)}Is St = T Detecting a decrease St = max(0, St-1 + (u - Xt - C)} Is St = T See ISYE-6501 Module 6(L1-L2) Playlist v0822 timestamp 5:05 / 5:49 in a regression model, given two variables A, B, what does it mean if the coefficient is a lot larger (and positive) for variable A (years of education), relative to variable B (age in years)