Advanced Machine Learning Deusto

In a multi-class classification where each class is equally relevant to the task, the dataset is fairly balanced which kind of metric is more appropriate to measure the performance of our model. Micro. Macro. Weighted.

In a multi-class classification where each class is equally relevant to the task, the dataset is not balanced which kind of metric is more appropriate to measure the performance of our model. Micro. Macro. Weighted.

In a multi-class classification where each class is not equally relevant to the task, the dataset is not balanced which kind of metric is more appropriate to measure the performance of our model. Micro. Macro. Weighted.

In a multi-class classification where each class is not equally relevant to the task, the dataset is balanced which kind of metric is more appropriate to measure the performance of our model. Micro. Macro. Weighted.

Active learning is located between... Supervised and Unsupervised. Classical ML and Deep ML. Heuristic and Fuzzy Logic.

In a scenario where there is a large dataset, but at least 70% of the data is actually unlabeled, which would be the most appropriate approach. Active Learning. Reinforcement Learning. Ensemble Learning.

In a multi-class classification, a multilayer-perceptron with a sigmoid function in the final layer follows a... One vs One strategy. One vs Rest strategy. None of Them.

In a multi-class classification, a multilayer-perceptron with a softmax function in the final layer follows a... One vs One strategy. One vs Rest strategy. None of Them.

In a multi-class classification, a SVC in the final layer follows a... One vs One strategy. One vs Rest strategy. None of Them.

In a classification context, you do not have information about the dataset of its distribution, or any related data, which one should apriori yield to a better result. Bagging. Pasting.

In a classification context, you do not have information about the dataset of its distribution, or any related data, which one should apriori yield to a better result. Bagging. Pasting. Boosting. Gradient Boosting.

Choose the correct one. AdaBoost is a boosting model based usually on several decision stomps. AdaBoost use the estimator weight to update the instance weight, which also depends in a predefined learning rate. A lower learning rate will induce the model in a local optima due to the sensibility to outliers.

Unlike AdaSyn, Smote focuses in all the minority classes. True. False.

Remove samples with the smallest average distance to closest minority neighbors. NearMiss-1. NearMiss-2. NearMiss-3. Tomek’s links.

Remove samples with the smallest average distance to farthest minority neighbors. NearMiss-1. NearMiss-2. NearMiss-3. Tomek’s links.

Remove pairs of very close instances, but of opposite classes. NearMiss-1. NearMiss-2. NearMiss-3. Tomek’s links.

In MCAR cases what is the best imputation method. Statistical Imputation. Univariate or multivariate Imputation. Missing value Indicator Imputation.

In MAR cases what is the best imputation method. Statistical Imputation. Univariate or multivariate Imputation. Missing value Indicator Imputation.

In MNAR cases what is the best imputation method. Statistical Imputation. Univariate or multivariate Imputation. Missing value Indicator Imputation.

Whats the null hypothesis of the ADF test?. there is a unit root present in the time series therefore the time series is not stationary. there is a unit root present in the time series therefore the time series is stationary. there is not a unit root present in the time series therefore the time series is stationary. there is not a unit root present in the time series therefore the time series is not stationary.

Whats the alternative hypothesis of the ADF test?. there is a unit root present in the time series therefore the time series is not stationary. there is a unit root present in the time series therefore the time series is stationary. there is not a unit root present in the time series therefore the time series is stationary. there is not a unit root present in the time series therefore the time series is not stationary.

What is a random walk?. Stationary + No correlated on the first difference. No stationary + No correlated on the first difference. Stationary + Correlated on the first difference.

Whats the null hyppthesis in Ljung-Box test?. data is independently distributed therefore no autocorrelation. data is not independently distributed therefore no autocorrelation. data is not independently distributed therefore autocorrelation. data is independently distributed therefore autocorrelation.

Which one is the null hypothesis in Granger causality test. y2,t does not Granger-cause y1,t (there is not Granger causality from time series 2 to time series 1). y2,t does Granger-cause y1,t (there is Granger causality from time series 2 to time series 1). y1,t does Granger-cause y2,t (there is Granger causality from time series 1 to time series 2). y1,t does not Granger-cause y2,t (there is no Granger causality from time series 1 to time series 2).