Machine learning

Question 1

Explain Parametric?

Answer 1

Parametric statistical procedures rely on assumptions about the shape of the distribution (i.e., assume a normal distribution) in the underlying population and about the form or parameters (i.e., means and standard deviations) of the assumed distribution. Advantage: Restrictive models are much more interpretive.

Question 2

Explain Non-parametric?

Answer 2

Nonparametric statistical procedures rely on no or few assumptions about the shape or parameters of the population distribution from which the sample was drawn. Advantage (compared to parametric methods): They may accurately fit a wider range of possible shapes for f. Disadvantage: A very large number of observations is required in order to obtain an accurate estimate of f

Question 3

Explain Supervised?

Answer 3

For each observation of the predictor measurement(s)𝑥i, 𝑖=1,…,𝑛 there is an associated response measurement 𝑦i
Supervised (examples):
• Linear regression
• Logistic regression
• Support vector machines
• Neural Networks
• Collaborative filtering (Methods that try to fill in the missing values e.g. Netflix ratings)

Question 4

What is Unsupervised?

Answer 4

We observe a vector of measurements 𝑥i, 𝑖=1,…,𝑛, but no associated response 𝑦i
Unsupervised (examples):
• Clustering
• PCA

Question 5

What is the Bias-variance tradeoff?

Answer 5

• We want a low variance and a low bias at the same time. However, when variance decreases, bias increases and vice versa.
• Bias refers to the error that is introduced by approxi- mating a real-life problem, which may be extremely complicated, by a much simpler model. For example, linear regression assumes that there is a linear relationship between Y and X1, X2, . . . , Xp. It is unlikely that any real-life problem truly has such a simple linear relationship, and so performing lin- ear regression will undoubtedly result in some bias in the estimate of f
• Variance is the amount that the estimate of the target function will change given different training data

Question 6

What is Quality of fit?

Answer 6

• There is no free lunch in statistics –> No one method dominates all others over all possible data sets
• Important task to decide for any given set of data which method produces the best results
• Selecting the best approach can be one of the most challenging parts of performing machine learning in practice
• In order to evaluate the performance of a machine learning method, we need to quantify the extent to which the predicted response value is close to the true response
• We compute the MSE by using our training data
• However, we are not interested whether our method works on the training data
• Rather we are interested how it works on our test data
• Suppose that we are interested in developing an algorithm to predict stock prices based on previous stock returns.
• We can train the method using stock returns from the past 6 months
• However, we are not interested in predicting last weeks stock return / price
• We are interested in predicting next weeks prices / returns
• How can we go about trying to select a method that minimizes the test MSE?
• In some settings, we may have a separate test data set (set of observations which we did not use to train the model)
• We can then simply evaluate our model on the test observations (compute the MSE of the test data)

Question 7

How can machine learning be included in your research?

Answer 7

• It could be used to figure out which consumers who use the deposit return system, and then based on that, it could be used to target advertising towards these consumers
• Used for prediction - would the people using it now also use it in two years? - long-term change
• Predict who wants to use it, based on previous users with similar characteristics, who are using it
• However, it might be quite hard, given that ML is usually based on big data, and the act of depositing bottles is a very analogue process, at least in Denmark, because you don’t have any technical touchpoints that connect the consumer to the behavior on an online data basis.
• A solution to this could be making the Deposit return system app based in the US.
• For that reason, it is even more important to do consumer research projects as ours on the topic, and even branch into even more descriptive data forms after – to be able to determine which consumers to target, and how