Pros and Cons of MSE
cost of interpretable
punish outliers more
minimazed by the mean
alternative Mean Absolute Error
indepent
feature
dependent
label
coefficient
weight
estimate
learn
prediction
inference
Statistic vs ML
Target of Interest
Structual and causal parameters
Prediction or Classification of Y
Focus
Formel properties (consistency & efficiency)
Does it work in practies (large and external data)
Model selection
researcher choses model based on princples
algorimthm estimate and compare alternative models
Judging qualtiy of estimation
Indirect (in-sample)
direct (out-of sample)
Inference
very important
less important
sample size
large N is nice
large N is required
No variable
much smaller than N
can be larger than N
preffered model complexity
often parametric/linear
complcated
Advantages of decision tree
many predictor variables (k>n)
high defree of non-lineartites
little distribuence due to irrelevant predictor variables
easy to explain and interepret the result
computionallly less demanding compared to RF
Advantages of random forest
often higher predictiv power than decision trees
more robust
Name five ethicial consideration in ML
Blackbox nature
Fairness and Non-discrimination
Training data sources
Manipulation of Feature
Energy Consumption
Zuletzt geändertvor 7 Monaten