Amazon Web Services · 2026 Edition
A complete preparation guide written by Amazon Web Services-certified engineers. Covers the exam format,all 4 blueprint domains, a week-by-week study plan, and proven tips for passing first time.
4–6 months
Prep time
Advanced
Difficulty
65
Exam questions
750/1000
Pass mark
Exam code
MLS-C01
Full name
AWS Machine Learning Specialty
Vendor
Amazon Web Services
Duration
180 minutes
Questions
65 items
Passing score
750/1000 (scaled)
Domains covered
4 blueprint domains
Recommended experience
1–2 years of ML/data science experience; Python proficiency; familiarity with ML frameworks (TensorFlow, PyTorch, scikit-learn)
Typical prep time
4–6 months
MLS-C01 earns the AWS Certified Machine Learning – Specialty designation. It validates the ability to design, implement, deploy, and maintain ML solutions on AWS using SageMaker and related services — the credential expected for ML engineering roles in cloud-native organisations.
Job roles this opens
Domain percentage weights are not currently available for this exam. The checklist below is still useful for planning your study.
Weeks 1–3
Data Engineering for ML: S3, Glue, Kinesis, data labelling (SageMaker Ground Truth)
Tip: SageMaker Ground Truth manages human labelling workflows for training data. Know the labelling job types: image classification, object detection, semantic segmentation, text classification, and NER. Know how active learning in Ground Truth reduces labelling cost by having the model label high-confidence examples automatically.
Weeks 4–6
Exploratory Data Analysis: feature engineering, encoding, scaling, class imbalance, bias detection
Tip: Class imbalance handling techniques are tested: oversampling minority class (SMOTE), undersampling majority class, adjusting class weights in the algorithm, or changing the decision threshold. Know when each technique is appropriate — SMOTE creates synthetic samples, not duplicates.
Weeks 7–9
Modelling: SageMaker built-in algorithms, custom containers, hyperparameter tuning, Autopilot
Tip: SageMaker built-in algorithms and their use cases: XGBoost (tabular regression/classification), K-Means (clustering), Random Cut Forest (anomaly detection), BlazingText (text classification, word2vec), DeepAR (time series forecasting), Object Detection, Image Classification, Seq2Seq. Questions give a use case and ask which algorithm fits.
Weeks 10–15
ML Implementation and Operations: SageMaker endpoints, batch transform, pipelines, monitoring, A/B testing
Tip: SageMaker Model Monitor is the primary tool for detecting model drift in production. Know the monitor types: Data Quality (detect input feature drift), Model Quality (detect prediction quality degradation), Bias Drift (detect fairness metric changes), and Feature Attribution Drift (detect SHAP value changes). Set up CloudWatch alarms on monitor metrics.
Amazon SageMaker Studio is the integrated ML development environment tested throughout MLS-C01. Know its components: Studio Notebooks (managed Jupyter), Experiments (track training runs), Model Registry (versioned model catalog), Pipelines (MLOps orchestration), and Model Monitor.
SageMaker training instance selection: GPU instances (ml.p3, ml.p4, ml.g4dn) for deep learning, CPU instances (ml.m5, ml.c5) for traditional ML algorithms like XGBoost. Know that distributed training (multiple GPU instances) requires a framework that supports data parallelism (Horovod, SageMaker Distributed Training).
Regularisation techniques to prevent overfitting are tested: L1 regularisation (Lasso, drives weights to zero, feature selection), L2 regularisation (Ridge, reduces weight magnitude, retains all features), Dropout (randomly disables neurons during training for neural networks). Know which technique is appropriate for which model type.
Amazon Comprehend (NLP), Amazon Rekognition (computer vision), Amazon Forecast (time series), and Amazon Personalize (recommendations) are managed AI services that appear in MLS-C01 as alternatives to building custom models on SageMaker. Know when to use a managed AI service vs a custom-trained SageMaker model.
Model evaluation metrics tested on MLS-C01: accuracy, precision, recall, F1 score, AUC-ROC, RMSE, MAE. Know which metric to use when: use F1 when class imbalance makes accuracy misleading, use AUC-ROC for binary classification with varying decision thresholds, use RMSE for regression when large errors should be penalised more than small ones.
Apply everything in this guide with adaptive practice questions, detailed answer explanations, and domain analytics.
Deep-dive explanations of the key topics tested on MLS-C01 — with exam key points and common misconceptions.