Machine Learning Interview Questions for 3 Years Experience
40 questions with answers for candidates with 3 years of experience — 2 easy, 30 medium, 8 hard.
What interviewers expect: 3 years experience
Three-year candidates are expected to have shipped models to production and handled issues after launch. Questions cover drift detection, monitoring, retraining pipelines, training-serving skew, latency optimisation and A/B testing. You may get an ML case study where you design the full solution and justify trade-offs between accuracy, cost and maintainability.
This set is weighted towards Decision Trees and Ensembles (7), Model Evaluation and Tuning (6), Regression Models (6), Data Preprocessing and Feature Engineering (6).
Production deployment Drift and monitoring A/B testing ML case studies
Solved at this levelStart practicing
3 QUESTIONS FREESign up to unlock the full question bank and track your progress.Unlock every question and answer, then continue practising company-wise and package-wise.
✓ 3 questions free • Unlock the rest in seconds
01
A new model has slightly better AUC but worse precision in the top 1,000 cases the team acts on. Which do you pick?
I would pick the model that does better where decisions are actually made. If the team only reviews the top 1,000 cases, precision at 1,000 or recall at that cut-off matters more than global AUC. I would make that the primary metric in future comparisons, and confirm the result holds across several time periods before switching.
02
How do you create polynomial features with regularisation in a pipeline?
from sklearn.preprocessing import PolynomialFeatures, StandardScaler
from sklearn.linear_model import Ridge
from sklearn.pipeline import make_pipeline
model = make_pipeline(
PolynomialFeatures(degree=2, include_bias=False),
StandardScaler(),
Ridge(alpha=1.0)
)
model.fit(X_train, y_train)
Tune degree and alpha together with GridSearchCV to avoid overfitting.
03
When would you use AdaBoost and how does it re-weight samples?
AdaBoost trains weak learners, usually decision stumps, in sequence. After each round it increases the weights of misclassified samples so the next learner focuses on them, and gives each learner a vote weight based on its accuracy. It works well on clean data but is sensitive to noisy labels and outliers, which get ever larger weights. Gradient boosting is a generalisation that has largely replaced it.
04
Lasso dropped a feature the business considers very important. How do you handle it?
Lasso often keeps only one of a group of correlated features, so the dropped feature may be redundant rather than useless. I would check correlations, try Elastic Net which keeps correlated groups, and compare cross-validated performance with the feature forced in. I would explain the trade-off clearly; if the accuracy difference is small, keeping the feature for trust and interpretability is reasonable.
The kernel trick computes dot products of data in a high-dimensional feature space without explicitly transforming the data, using a kernel function such as RBF, exp(-gamma times squared distance). This lets linear algorithms like SVM learn non-linear boundaries efficiently. In RBF SVMs, larger gamma makes each point's influence more local, which can lead to overfitting.
06
Your training loss is not decreasing at all. What could be wrong?
Common causes are a learning rate that is too high or too low, unscaled features, a bug in labels or feature alignment, wrong loss for the task, or data that has been shuffled away from its labels. I would try overfitting a tiny subset first; if the model cannot even fit 50 rows, there is a bug. Then check scaling, learning rate and gradient values.
A model registry is a central store of model versions with metadata: training data version, parameters, metrics, owner and stage such as staging or production. It makes promotions and rollbacks controlled and auditable, and lets serving systems load a model by name and version. MLflow Model Registry, SageMaker Model Registry and Vertex AI Model Registry are common examples.
First understand why values are missing: completely at random, at random, or not at random. Options are dropping rows or columns with heavy missingness, simple imputation with mean, median or mode, model-based imputation like KNNImputer or IterativeImputer, or a separate 'missing' category. Adding a missing-indicator column often helps because missingness itself can be predictive, such as income left blank.
09
A product team wants 'similar products' recommendations with no purchase labels. How would you start?
I would build product embeddings from content: TF-IDF or sentence embeddings of titles and descriptions, plus category and price features. Then use cosine similarity with nearest-neighbour search, such as scikit-learn NearestNeighbors or an approximate index like FAISS for large catalogues. I would review samples with the product team and later add co-view or co-purchase signals once collected.
10
What is the difference between offline evaluation and online A/B testing of a model?
Offline evaluation measures metrics like AUC on historical holdout data; it is fast and cheap but cannot capture user behaviour changes or feedback loops. Online A/B testing exposes a share of real traffic to the new model and measures business outcomes like conversion or loss rate. Offline gains do not always translate online, so important models are validated with A/B tests or shadow deployments.
11
Why do skewed features often need a log or power transform?
Heavily right-skewed features like income or transaction amount have long tails that can dominate linear models and distance metrics. A log1p, Box-Cox or Yeo-Johnson transform compresses large values and makes the distribution closer to normal, which often improves linear and distance-based models. Yeo-Johnson works with zero and negative values. Tree models usually do not need these transforms.
12
What is target encoding and how do you avoid leakage with it?
Target encoding replaces each category with the mean target value for that category, which works well for high-cardinality features. It leaks if computed on the full data because each row sees its own label. Avoid this by computing encodings out-of-fold within cross-validation and adding smoothing toward the global mean for rare categories. scikit-learn 1.3+ has TargetEncoder with built-in cross-fitting.
13
You are asked to predict customer churn. How would you frame the ML problem from scratch?
First define churn precisely, for example no purchase in 60 days, and the prediction window. It becomes binary classification. Build features from data available before the cut-off date only, choose a time-based split, and pick metrics like recall at a fixed precision or lift in the top decile. Start with logistic regression as a baseline, then try boosting, and agree on how retention teams will use the scores.
14
Your manager asks for a single number showing the model's business value. What do you give?
I would convert model outcomes into money or time using a cost matrix agreed with the business, for example savings from caught fraud minus the cost of reviewing false alerts. I would compare this against the current process or a rule-based baseline on holdout data and show a range rather than one point. A short note on assumptions keeps the number honest.
Agglomerative hierarchical clustering starts with each point as its own cluster and repeatedly merges the two closest clusters. Linkage defines closeness: single, complete, average or Ward, which minimises variance. A dendrogram shows the merge history, and cutting it at a height gives the clusters. It does not need k upfront but scales poorly, roughly O(n squared) memory, so it suits smaller datasets.
16
What should you monitor for a model in production?
Monitor service health such as latency, error rate and throughput; input data quality such as missing values, schema changes and out-of-range values; feature and prediction distribution drift; and real performance once labels arrive. Also track business KPIs the model influences. Set alerts with thresholds and owners, and keep a dashboard so issues are found before users report them.
17
Two models have RMSE 120 and 125, but the second has much lower MAE. Which do you choose?
It depends on error cost. The first model makes fewer big errors, while the second is better on typical cases but has some large misses. If large errors are expensive, such as stock-outs of costly items, choose the first. If typical accuracy matters more, choose the second. I would also check error distributions by segment and discuss with stakeholders before deciding.
Log loss, or cross-entropy, is the negative average log of the probability assigned to the true class. It heavily penalises confident wrong predictions, such as predicting 0.99 for an event that does not happen. It evaluates probability quality, not just ranking, so it suits problems where probabilities drive decisions. Lower is better, and a constant base-rate prediction is the baseline to beat.
19
How do one-vs-rest and multinomial approaches differ for multiclass classification?
One-vs-rest trains one binary classifier per class against all others and picks the highest score; it is simple but probabilities may not sum to one. Multinomial (softmax) logistic regression models all classes jointly with a single loss, giving consistent probabilities. For metrics, use macro averaging to treat classes equally or weighted averaging to account for class frequency.
Isolation Forest builds random trees that split on random features and random thresholds. Anomalies are few and different, so they get isolated in fewer splits, giving shorter average path lengths. The anomaly score comes from that path length. It is fast, works without labels and scales well. The contamination parameter sets the expected share of anomalies and determines the threshold.
21
Describe a project where you improved a model's performance significantly.
Situation: a product-return prediction model had 0.68 AUC with logistic regression. Task: improve it for a warehouse pilot. Action: I engineered features like past return rate per customer and category, moved to LightGBM with early stopping, and tuned it with Optuna on time-based folds. Result: AUC rose to 0.81, and the pilot reduced avoidable returns in two categories. I documented each step's gain.
Use statsmodels on the numeric feature matrix with a constant added.
import pandas as pd
from statsmodels.stats.outliers_influence import variance_inflation_factor
from statsmodels.tools import add_constant
Xc = add_constant(X_num)
vif = pd.Series([variance_inflation_factor(Xc.values, i)
for i in range(Xc.shape[1])], index=Xc.columns)
print(vif.drop('const'))
Values above 5 to 10 indicate strong multicollinearity.
23
When should you use TimeSeriesSplit or GroupKFold instead of normal k-fold?
Use TimeSeriesSplit when data is ordered in time, so each validation fold comes after its training data and no future information leaks back. Use GroupKFold when multiple rows belong to the same entity, such as several visits per patient, so the same entity never appears in both train and validation. Normal k-fold in these cases gives over-optimistic scores.
24
Compare filter, wrapper and embedded feature selection methods.
Filter methods score features independently of a model using statistics such as correlation, chi-square or mutual information; they are fast but ignore interactions. Wrapper methods like Recursive Feature Elimination train models on feature subsets; they are accurate but slow. Embedded methods select during training, such as L1 regularisation or tree feature importances. Embedded methods are a common practical middle ground.
25
Train a random forest and show feature importances in scikit-learn.
Profile first to see whether time goes to feature lookup, preprocessing or the model. Then cache or precompute features, load the model once at startup, reduce trees or features, use a faster library like LightGBM or ONNX Runtime, and batch requests if possible. Run several workers and keep the service close to its data store. Re-measure p95 and p99 latency, not just the average.
27
Use HistGradientBoosting with native categorical support in scikit-learn.
from sklearn.ensemble import HistGradientBoostingClassifier
clf = HistGradientBoostingClassifier(
categorical_features='from_dtype', max_iter=500,
learning_rate=0.05, early_stopping=True, random_state=42)
X_train[cat_cols] = X_train[cat_cols].astype('category')
clf.fit(X_train, y_train)
It handles missing values natively and is much faster than GradientBoostingClassifier on large data.
As the number of features grows, data becomes sparse in the feature space, distances between points become similar, and the amount of data needed to generalise grows rapidly. Distance-based methods like k-NN and k-means suffer most. Remedies include feature selection, dimensionality reduction such as PCA, regularisation and collecting more data. Text and genomics data often face this problem.
29
XGBoost validation loss starts rising after 300 rounds while training loss keeps falling. What do you do?
That is overfitting beyond about 300 rounds. I would use early stopping with a patience such as 50 rounds on a validation set so training stops at the best iteration, then lower the learning rate, reduce max_depth, add subsample and colsample_bytree below 1, and increase min_child_weight or lambda. The final model is refit using the best iteration count found.
30
How do you handle pressure to report better model numbers than the data supports?
I explain the risk clearly: inflated numbers lead to bad decisions and lose trust when production results arrive. I present the honest metric with confidence intervals, and then offer a concrete plan to improve it, such as new data sources. In one case I showed that a 'better' number came from leakage, and leadership agreed to report the lower but reliable figure.
31
What is regularisation in general and why does it reduce overfitting?
Regularisation adds a constraint or penalty that discourages overly complex models, such as large weights in linear models, deep trees or too many boosting rounds. It trades a little extra bias for a larger drop in variance, so test error falls. Examples include L1 and L2 penalties, dropout in neural networks, early stopping, pruning and limiting tree depth.
Data drift, or covariate shift, means the distribution of input features changes, such as average order value rising after a pricing change. Concept drift means the relationship between features and target changes, such as fraud patterns evolving. Data drift can be detected without labels using PSI or KS tests; concept drift usually needs labels and shows up as falling performance. Both trigger investigation and retraining.
33
A demand forecast is consistently under-predicting during festivals. How would you fix it?
The model lacks festival information. I would add features for festival dates such as Diwali or regional holidays, days-to-festival, and past festival uplift for the same item. If the business cost of stock-outs is higher than over-stock, I would use a quantile loss for a higher percentile. Then validate specifically on past festival periods, not just the average error.
34
Tell me about a time you had to explain a model's error to a business team.
Situation: a sales forecast missed by 18 percent in one month. Task: explain why to the category team. Action: I broke the error down by region and found most came from a sudden promotion not in our data. I showed a simple chart and proposed adding a promotion calendar feed. Result: the team shared promotion plans in advance, and the next quarter's error fell below 8 percent.
35
Your preprocessing works in notebooks but gives different results in production. Why?
Common causes are preprocessing written separately in two places, statistics like means recomputed on production data instead of using training values, different library versions, or different column order and data types. The fix is to save one fitted Pipeline with preprocessing and model together, pin versions, validate input schema, and test that the saved pipeline gives identical outputs on sample rows.
36
Your classifier works well overall but poorly for one customer segment. What do you do?
I would measure metrics per segment to confirm, then check if that segment is under-represented or has different feature distributions. Fixes include collecting more data, sample weighting, segment-specific features, or a separate model if the behaviour is truly different. If the segment relates to a protected attribute, I would also review fairness metrics and involve the compliance team.
37
How do you build a text classifier with TF-IDF and logistic regression?
Reduce size by limiting max_depth and min_samples_leaf, using fewer trees since gains flatten after a few hundred, and dropping weak features. Alternatively switch to a gradient boosting model like LightGBM, which usually gives similar accuracy with far smaller, faster models. Converting to ONNX or using a compiled predictor can also speed up inference. Measure accuracy and latency after each change.
39
What makes XGBoost, LightGBM and CatBoost different?
XGBoost adds regularisation to the objective, uses second-order gradients and handles missing values natively. LightGBM grows trees leaf-wise and uses histogram binning, making it very fast on large data, though it can overfit small data. CatBoost handles categorical features with ordered target statistics and uses symmetric trees, giving strong defaults with less tuning. All three support GPU training.
40
Tell me about a time you took a model from notebook to production.
Situation: our demand model ran only in a notebook on my laptop. Task: make it run daily for 300 stores. Action: I refactored it into modules with a scikit-learn pipeline, added tests and input checks, containerised it with Docker and scheduled it with Airflow writing forecasts to the warehouse. Result: forecasts became available by 6 a.m. daily without manual work, and failures triggered alerts instead of silent gaps.
Frequently asked questions
What is asked in Machine Learning interviews for 3 years of experience?
Three-year candidates are expected to have shipped models to production and handled issues after launch. Questions cover drift detection, monitoring, retraining pipelines, training-serving skew, latency optimisation and A/B testing. You may get an ML case study where you design the full solution and justify trade-offs between accuracy, cost and maintainability.
How many questions should I practise before a Machine Learning interview?
Work through all 40 questions here, then the company-wise sets for the employers you are targeting. Quality matters more than count: be able to explain why each answer is right.
Which topics matter most for 3 years of experience?
Focus first on Production deployment, Drift and monitoring, A/B testing, ML case studies.