quaxp-1-final-1
You are in possession of a dataset containing individual reports of UFO sighting. Here is a list of the attributes in the dataset: place each attribute to its type.
Year
Month
city
state
country
shape
duration (seconds)
duration (hours/min)
comments
date posted
latitude
longitude
quaxp-1-final-2
You keep working on the same dataset. If you were to predict each of these attributes, which type of task would you choose?
Year
Month
state
country
shape
duration (seconds)
latitude
longitude
Select the true statements about this histogram:
- There are approximately 60000 instances with a value comprised between 0 and 500.
- There are approximately 3000 instances with a value over 1000.
- There are approximately 10000 instances with a value under 1000.
- There are approximately 2000 instances with a value over 3300.
Select the true statements about this boxplot:
- 50% of the values for Year in the dataset are under 1990.
- 25% of the values for Year in the dataset are above 2000.
- 10% of the values for Year in the dataset are under 1985.
- Most of the values for Year in the dataset are above 2000.
In the UFO dataset, the attribute country contains 11791 missing values, on a total of 87512 instances (13% of missing values). What can you do?
- Set up a neutral category for the missing values of country
- Recover the values from the attributes city and state
- Recover the values from the attributes latitude and longitude
- Drop the attribute
- Drop the rows
The attribute “duration (seconds)” contains 7037 wrong values (values filled with 0). The attribute “duration (hours/min)” looks like the picture above. Is it possible to recover the values of the attribute “duration (seconds)”? You want to build a model to predict the shape of the UFO. Which statements are true?
- It would be possible to recover the values from the attribute “duration (hours/min)” with an advanced technique of text analysis.
- We can drop the attribute.
- We can drop the instances with a wrong value.
- We need to make some further analysis on the predictive model before taking a decision about dropping the instances or the attribute.
- We can ignore the problem and keep the wrong values of 0.
You want to build a model to predict the shape of a UFO. Which attributes do you keep for the prediction?
- Year
- Month
- city
- state
- country
- duration (seconds)
- duration (hours/min)
- comments
- date posted
- latitude
- longitude
We are now in possession of a dataset containing information about some wines. The goal is to predict the attribute “points” (which is a grade for the wine in percent, from 0 to 100) from its price. This graph is the results of the prediction vs. true label.
Which statements are true about the graph?
- The model’s predictions are quite on point.
- The predictions range of the model is much smaller than the real values.
- The model doesn’t seem to have found a good prediction pattern.
- The predictions range of the model is as big as the one of the real data.
With the same wine problem as in the previous question, here is the description of the two attributes (price is the predictive attribute, points is the predicted attribute).
Would a MAE of 2,25 (for the attribute points) be considered good?
This is the list of the attributes in another wine dataset. All the attributes are numerical, except for the categorical attribute “type”, which is composed of the labels “red” and “white”.
We want to predict the attribute “quality” (a grade between 0 and 10), based on the other attributes. On which attribute could we split the dataset to try to increase the performance of the model?
- fixed acidity
- volatile acidity
- citric acid
- residual sugar
- chlorides
- free sulfur dioxide
- total sulfur dioxide
- density
- pH
- sulphates
- alcohol
- type