SAS statistical certifications Statistical & Machine Learning for Agriculture: practice questions
SAS statistical certifications (Statistical Business Analyst, Applied Statistics Associate, Statistics for Machine Learning) · USA / global · SAS Institute
Original practice questions written to the published syllabus · not past papers
10worked here, free
93written to this exam in this subject
11syllabus units
About SAS statistical certifications
SAS statistical certifications (Statistical Business Analyst, Applied Statistics Associate, Statistics for Machine Learning) is a professional certification run by SAS Institute, for candidates worldwide. Proctored online or test-centre exams per credential.
Checked against the exam body's published information. Formats and fees change; confirm the current edition there before you book.
Statistical & Machine Learning for Agriculture in SAS statistical certifications
Predictive modelling workflow: partitioning, regularisation, trees and ensembles, neural networks, model assessment and interpretation.
Prediction from data with honest validation: the bias-variance trade-off, regularised regression, trees and ensembles, support vector machines, neural networks and deep learning for images, unsupervised learning, and how to use them on yield, disease, price and image data without fooling yourself. The Agribusiness and Agricultural Engineering tracks introduce ML; this is the statistician's full course.
Our 93 SAS statistical certifications questions in this subject: 26 easy, 52 medium, 15 hard, across 11 units.
10 practice questions, worked
Choose an answer, then open the working. Each explanation says why the right answer is right and why the others are not.
These 10 were written to SAS statistical certifications and are not shown on any other exam's page.
Question 1 · medium
What does the one-standard-error rule select in cross-validation?
The most complex model that converges
The simplest model within one SE of the minimum error
The model with the smallest training error
The model with error exactly one SE above the minimum
Answer and working
B. The simplest model within one SE of the minimum error
It favours parsimony: among models whose CV error is statistically indistinguishable from the best, choose the simplest.
Question 2 · medium
A model to predict yields in new districts is validated by randomly assigning district-year records to folds. What does this ignore?
The need to standardise yields before fitting
Spatial dependence; hold out district blocks
The need to use leave-one-out instead of k-fold
The class imbalance between high- and low-yield districts
Answer and working
B. Spatial dependence; hold out district blocks
Neighbouring districts and the same district in other years share conditions; random folds put near-copies in training and test sets. Holding out whole spatial blocks matches the goal of predicting new districts.
Question 3 · easy
Which validation design matches the aim of forecasting next season's yields?
Holding out one random field from each year
Testing on the same years used for training
Training on earlier years and testing on later years
Randomly shuffling all years into folds
Answer and working
C. Training on earlier years and testing on later years
Forward-chaining (or leave-one-year-out) validation reproduces the forecasting situation, where only past seasons are available.
Question 4 · medium
Why does ridge regression have a unique solution even when there are more predictors than observations?
X'X + lambda I is invertible for lambda > 0
Ridge regression removes predictors until p < n
The ridge solution ignores the predictor matrix
Least squares is always unique for any p
Answer and working
A. X'X + lambda I is invertible for lambda > 0
Adding lambda I makes the matrix positive definite, so beta-hat = (X'X + lambda I)^(-1) X'y exists; ordinary least squares has no unique solution when p > n.
Question 5 · medium
A logistic model for blast infection has intercept -2.0 and coefficient 0.5 per day of leaf wetness. What is the predicted probability of infection after 6 days of wetness?
0.500
0.731
2.718
1.000
Answer and working
B. 0.731
Log-odds = -2 + 0.5 x 6 = 1; p = 1/(1 + e^-1) = 0.731. 1.000 is the log-odds; 2.718 is the odds; 0.500 corresponds to log-odds 0.
Question 6 · easy
How does a random forest differ from bagged trees?
Each tree is fitted to the residuals of the previous tree
Each split considers only a random subset of predictors
Trees are fitted without bootstrap samples
Each tree is pruned to a single split
Answer and working
B. Each split considers only a random subset of predictors
Random predictor subsets decorrelate the trees, so averaging reduces variance more; fitting residuals sequentially is boosting.
Question 7 · medium
In gradient boosting with squared-error loss, what is each new tree fitted to?
The predictions of the previous tree
The current residuals
The original outcome values
A bootstrap sample of the outcome
Answer and working
B. The current residuals
With squared error the negative gradient is the residual, so each small tree corrects what the current ensemble gets wrong; it is added after scaling by the learning rate.
Question 8 · medium
What is the kernel trick?
Replacing the hinge loss by squared error
Choosing C by cross-validation automatically
Removing outliers before fitting the SVM
Inner products in feature space, implicitly
Answer and working
D. Inner products in feature space, implicitly
Because the SVM solution depends only on inner products, replacing them by a kernel fits a linear boundary in an implicit high-dimensional space.
Question 9 · easy
Three farms assigned to a k-means cluster have (standardised) coordinates (2, 4), (4, 6) and (6, 8). What is the updated cluster centroid?
(2, 4)
(4, 6)
(12, 18)
(6, 8)
Answer and working
B. (4, 6)
The centroid is the mean of member coordinates: ((2 + 4 + 6)/3, (4 + 6 + 8)/3) = (4, 6). (12, 18) is the sum.
Question 10 · medium
What is a valid conclusion from a t-SNE plot of hyperspectral pixels?
The axes are interpretable as principal components
Points close together are similar
Larger clusters contain more variable pixels
The distance between clusters measures their dissimilarity
Answer and working
B. Points close together are similar
t-SNE preserves local neighbourhoods but distorts global distances and cluster sizes, so only local closeness should be read from it.
A study plan for Statistical & Machine Learning for Agriculture in SAS statistical certifications
This plan follows where our questions actually fall, so the time goes where the marks are likely to be. It is our reading of the syllabus, not the exam body's weighting.
First, the heavy units (40% of the 93 questions): Neural networks and deep learning; CNNs for plant and field images (10, 11%); Cross-validation: k-fold and leave-one-out; spatial and temporal blocking (9, 10%); Decision trees, random forests and gradient boosting (9, 10%); Linear and quadratic discriminant analysis and logistic classification (9, 10%).
Read the manual chapters for these first and do every worked question you can find on them.
Then the middle band: Dimensionality reduction: PCA and embeddings (8, 9%); k-nearest neighbours and naive Bayes (8, 9%); Model interpretation: variable importance and SHAP (8, 9%); Regularised regression: ridge, lasso and elastic net (8, 9%).
Aim for one sitting per unit, then mixed questions across them.
Then the lighter units: Supervised learning and the bias-variance trade-off (8, 9%); Support vector machines (8, 9%); Unsupervised learning: k-means, k-medoids and hierarchical clustering (8, 9%).
Learn the definitions, classifications and standard numbers; these units are more often recalled than worked.
Leave time for the hard ones: k-nearest neighbours and naive Bayes (2 hard of 8), Model interpretation: variable importance and SHAP (2 hard of 8), Regularised regression: ridge, lasso and elastic net (2 hard of 8).
Come back to these after a first pass, untimed, and read every explanation, including for the options you rejected.
Bank the easy marks: k-nearest neighbours and naive Bayes (3 easy of 8); Regularised regression: ridge, lasso and elastic net (3 easy of 8).
Do these early in revision so they are fast on the day.
Finish timed. Sit the subject as a timed set without notes, mark it, and send each wrong answer back to its unit above.