Free Practice Questions for KBI 금융 AI 리터러시 (K-ALFA)

Exam-style questions and explanations by OpenExamPrep.

✓ No registration✓ No credit card

100+ Questions

100% Free

Loading practice questions...

Exam Review

Key Facts: KBI 금융 AI 리터러시 (K-ALFA) Exam

KRW 55,000

Official K-ALFA exam fee (전과목)

KBI 공개시험형 table, 등록번호 2025-005556 (checked 2026-09-20)

90 minutes

Official examination duration

KBI K-ALFA qualification page (checked 2026-09-20)

100 points

Official scoring scale; item count not published

KBI K-ALFA qualification page (checked 2026-09-20)

60 points

Passing score on the 100-point scale

KBI K-ALFA qualification page (checked 2026-09-20)

Four-option MCQ

Official objective format (객관식 4지선다)

KBI K-ALFA qualification page (checked 2026-09-20)

The KBI 금융 AI 리터러시 (K-ALFA) is a 90-minute, 80-question CBT assessment administered by the Korea Banking Institute testing AI fundamentals, financial industry applications, and regulatory governance. With a 60% passing mark, this 100-question study bank offers dual-language Korean/English technical terminology across ML algorithms, financial data preprocessing, credit scoring, FDS fraud detection, SHAP/LIME explainability, and FSC AI compliance.

Sample KBI 금융 AI 리터러시 (K-ALFA) Practice Questions

Try these sample questions to review concepts for the KBI 금융 AI 리터러시 (K-ALFA) exam. Each question includes a detailed explanation. Start the interactive quiz above for the full 100+ question experience with AI tutoring.

1Which of the following correctly describes the hierarchical relationship between Artificial Intelligence (인공지능), Machine Learning (머신러닝), and Deep Learning (딥러닝)?

A.Artificial Intelligence, Machine Learning, and Deep Learning are entirely disjoint, mutually exclusive paradigms utilized for different asset classes in finance.

B.Machine Learning is the overarching field that encompasses Artificial Intelligence, while Deep Learning is an independent methodology unrelated to neural networks.

C.Deep Learning is the broadest domain encompassing Machine Learning, which in turn includes Artificial Intelligence as a rule-based expert system.

D.Artificial Intelligence is the broadest concept, Machine Learning is a subset of AI, and Deep Learning is a specialized subset of Machine Learning based on multi-layered neural networks.

Explanation: Artificial Intelligence (인공지능) is the broadest field encompassing machines performing human-like cognitive tasks. Machine Learning (머신러닝) is a subfield of AI that learns patterns from data without explicit programming, and Deep Learning (딥러닝) is a specialized subset of ML utilizing artificial neural networks with multiple hidden layers.

2Which of the following financial use cases best illustrates a Supervised Learning (지도학습) classification problem?

A.Grouping banking customers into five distinct marketing clusters without pre-existing demographic labels

B.Predicting whether a corporate loan applicant will default (1) or repay (0) based on historical labeled financial statements

C.Detecting novel, unseen zero-day malware patterns on banking servers without training labels

D.Training an autonomous trading agent to execute limit orders through trial-and-error reward signals

Explanation: Supervised Learning (지도학습) requires labeled training data comprising input features and ground-truth target labels. Binary classification of loan default (부도/정상) based on historical repayment records is a classic supervised learning task.

3What is the primary characteristic of Unsupervised Learning (비지도학습) in banking operations?

A.It requires human annotators to manually label every transaction with fraudulent or non-fraudulent tags before training.

B.It discovers hidden patterns, underlying structures, or natural clusterings in datasets without predefined target labels.

C.It relies exclusively on environment state transitions and scalar numerical penalties to update model weights.

D.It is restricted strictly to continuous numerical regression forecasting such as interest rate projection.

Explanation: Unsupervised Learning (비지도학습) operates on unlabeled data (라벨이 없는 데이터) to uncover intrinsic data distributions, correlations, or clusters. In banking, it is widely utilized for Anti-Money Laundering (AML) anomaly detection and customer behavior grouping.

4In Reinforcement Learning (강화학습), an autonomous agent learns optimal policies through interactions with an environment. In the context of algorithmic execution (알고리즘 매매), what does the 'reward' (보상) typically represent?

A.The pre-labeled ground truth classification category assigned by financial regulators

B.The Euclidean distance between customer bank account balances in vector space

C.A quantitative performance feedback signal, such as realized profit-and-loss (P&L) or execution slippage minimization

D.The total number of hidden layers configured inside a convolutional neural network

Explanation: In Reinforcement Learning (강화학습), the reward signal (보상 함수) provides scalar feedback guiding the agent toward optimal actions. In trade execution algorithms, positive rewards correspond to minimal market impact, low execution slippage, or maximized risk-adjusted returns.

5When constructing a Decision Tree (의사결정나무) for retail loan default prediction, which splitting criterion measures node impurity based on class probability distribution?

A.The Population Stability Index (PSI) measuring long-term covariate shift

B.Root Mean Squared Error (RMSE) applied to discrete categorical classifications

C.The Sharpe Ratio computed across historical financial market returns

D.Gini Impurity (지니 불순도) or Information Gain / Entropy (정보 획득량 / 엔트로피)

Explanation: Decision Trees split child nodes using impurity metrics such as Gini Impurity (CART algorithm) or Entropy / Information Gain (ID3, C4.5). These metrics evaluate how homogenous the classes (e.g., default vs. non-default) are within each partition.

6What is the primary algorithmic difference between Bagging (배깅) and Boosting (부스팅) in ensemble machine learning?

A.Bagging trains multiple base models sequentially to correct previous residuals, while Boosting trains models in parallel on independent random samples.

B.Bagging trains multiple independent models in parallel using bootstrap aggregation to reduce variance, whereas Boosting trains models sequentially to reduce bias by focusing on previous errors.

C.Bagging is applied exclusively to deep neural networks, while Boosting can only be applied to linear regression models.

D.Bagging completely eliminates model bias, whereas Boosting is incapable of handling tabular credit datasets.

Explanation: Bagging (Bootstrap Aggregating, e.g., Random Forest) trains base learners in parallel on bootstrap samples to reduce model variance. Boosting (e.g., XGBoost, LightGBM) trains learners sequentially, with each subsequent model emphasizing the errors (residuals) of previous estimators to reduce bias.

7Why are Gradient Boosted Decision Tree (GBDT) frameworks like LightGBM and XGBoost frequently preferred over deep neural networks for institutional tabular credit scoring (CSS)?

A.GBDT algorithms require perfectly stationary continuous time-series data and cannot accept categorical inputs.

B.GBDT models efficiently handle mixed tabular data types, missing values, and non-linear feature interactions without extensive manual scaling, yielding superior out-of-the-box performance on tabular data.

C.GBDT frameworks guarantee zero mathematical bias and produce fully explainable mathematical closed-form equations identical to linear regression.

D.Financial regulators explicitly prohibit the use of deep neural networks in commercial banks while mandating XGBoost for all retail lending.

Explanation: Tabular datasets in banking feature diverse numerical distributions, categorical variables, and missing entries. Tree-based gradient boosting models (XGBoost, LightGBM, CatBoost) are invariant to monotonic transformations, robust to unscaled features, and consistently outperform deep networks on structured tabular credit benchmarks.

8In financial Fraud Detection Systems (FDS), fraudulent transactions typically comprise less than 0.1% of all records (extreme class imbalance / 극심한 클래스 불균형). Why is standard Accuracy (정확도) an inappropriate evaluation metric in this scenario?

A.Accuracy cannot be calculated on binary classification models containing more than 10,000 observations.

B.A trivial baseline model that classifies every transaction as legitimate would achieve 99.9% accuracy while failing to detect a single instance of fraud.

C.Accuracy mathematically penalizes false positives twice as heavily as false negatives in transaction monitoring.

D.Financial Supervisory Service (금감원) regulations require that accuracy be replaced exclusively by the Pearson correlation coefficient.

Explanation: Under severe class imbalance (클래스 불균형), a naive majority-class classifier predicting 'legitimate' for every transaction attains 99.9% accuracy despite 0% fraud recall. Consequently, practitioners rely on Precision-Recall AUC (PR-AUC), F1-Score, and Recall (재현율) to evaluate fraud detection efficacy.

9How does the Synthetic Minority Over-sampling Technique (SMOTE / 합성 소수 오버샘플링) address class imbalance in credit default datasets?

A.It randomly removes majority-class instances until both classes contain identical sample counts.

B.It synthesizes new minority-class samples along line segments connecting existing minority instances and their k-nearest neighbors in feature space.

C.It applies L1 Lasso regularization to shrink the coefficients of non-defaulting borrowers to zero.

D.It converts the binary classification task into an unsupervised principal component clustering problem.

Explanation: SMOTE (합성 소수 오버샘플링 기법) interpolates synthetic minority-class observations along the line segments connecting minority instances to their k-nearest neighbors in feature space. This prevents simple duplicate overfitting associated with naive random oversampling.

10In quantitative portfolio risk modeling, what does the Bias-Variance Tradeoff (편향-분산 트레이드오프) describe?

A.High bias leads to underfitting by oversimplifying model assumptions, while high variance leads to overfitting by modeling noise in training data.

B.High bias causes the model to fit every random noise fluctuation, while high variance results in linear underfitting.

C.Bias and variance both decrease simultaneously to zero as model parameter complexity increases toward infinity.

D.The tradeoff is only relevant to physical robotics and has no mathematical bearing on financial time-series forecasting.

Explanation: The Bias-Variance Tradeoff represents the balance between systematic model simplification errors (high bias / underfitting, 과소적합) and excessive sensitivity to training set fluctuations (high variance / overfitting, 과적합). Optimal financial generalization requires minimizing total expected prediction error.

About the KBI 금융 AI 리터러시 (K-ALFA) Exam

The Korea Financial AI Literacy (KBI 금융 AI 리터러시 / K-ALFA) examination is South Korea's premier professional credential testing foundational artificial intelligence knowledge, practical banking implementation capabilities, and regulatory compliance under the Financial Services Commission (FSC) AI Guidelines. This 100-question practice bank delivers comprehensive coverage of AI/ML algorithms, financial data engineering, robo-advisory, alternative credit scoring, fraud detection, explainable AI (XAI), and financial cybersecurity.

Exam sponsor: Korea Banking Institute (한국금융연수원 / KBI). The requirements and fees below concern the certification or admission exam, separate from our free practice resources.

Assessment

Question count not published by the exam provider

Time Limit

90 minutes

Passing Score

60 of 100 points (60%)

Exam / Certification Fees

KRW 55,000

Exam sponsor website

Fees, eligibility, and exam policies can change. Confirm them with the exam sponsor before applying or paying.

Official sources

Our practice resources: topics covered

We aim to reflect publicly available exam outlines and topic information in our study resources. Coverage, format, and difficulty may differ from the actual exam, and we cannot guarantee that every detail is accurate or current. Confirm exam requirements, fees, and policies with the official exam sponsor.

40 of 100 questions

AI Fundamentals & Financial Data (인공지능 핵심 원리 및 금융 데이터)

Core concepts of artificial intelligence, machine learning algorithms (supervised, unsupervised, reinforcement learning), deep learning architectures (CNN, RNN/LSTM, Transformers), Large Language Models (LLMs), financial data characteristics, structured vs. unstructured data, data preprocessing, feature engineering, synthetic financial data, vector embeddings, and Retrieval-Augmented Generation (RAG) in financial services.

25 of 100 questions

Financial AI Applications (금융 인공지능 실무 응용)

Practical AI implementations across banking and capital markets: robo-advisors, algorithmic asset allocation, alternative-data credit scoring (CSS), AI-driven Fraud Detection Systems (FDS), conversational AI chatbots, automated underwriting, OCR document processing, and financial sentiment analysis.

35 of 100 questions

AI Ethics, Regulation & Governance (금융 AI 윤리·규제 및 거버넌스)

Financial AI ethics, fairness and bias mitigation, Explainable AI (XAI: SHAP, LIME), privacy-preserving machine learning (differential privacy, federated learning, pseudonymization under the Three Data Acts), Financial Services Commission (FSC) AI Guidelines in Finance (금융분야 AI 가이드라인), EU AI Act overview, AI risk management frameworks, model validation, and AI cybersecurity (prompt injection, poisoning, model inversion).

Preparing for the KBI 금융 AI 리터러시 (K-ALFA) Exam

What You Need to Know

Passing score: 60 of 100 points (60%)
Assessment: Question count not published by the exam provider
Time limit: 90 minutes
Exam / certification fees: KRW 55,000 Official sources

Using Our Practice Resources

Work through all 100 available questions
Review every answer and explanation
Track weak areas and revisit them
Use our AI tutor for tough concepts

KBI 금융 AI 리터러시 (K-ALFA): Suggested Study Strategy

1Differentiate Supervised vs. Unsupervised vs. Reinforcement Learning: Understand that credit default prediction uses supervised classification, transaction clustering for AML uses unsupervised anomaly detection, and algorithmic execution or market making uses reinforcement learning.

2Master Imbalanced Data Handling in Finance: In fraud detection and loan default prediction, positive cases are extremely rare. Focus on techniques such as SMOTE (Synthetic Minority Over-sampling Technique), focal loss, cost-sensitive learning, and PR-AUC evaluation rather than raw accuracy.

3Understand RAG Pipeline Mechanics: Learn how dense vector embeddings, cosine similarity search in vector databases (e.g., Milvus, Pinecone), and context-injected LLM prompts reduce financial hallucination in internal policy retrieval.

4Differentiate Global vs. Local Explainable AI (XAI): SHAP (Shapley Additive exPlanations) is rooted in cooperative game theory and provides both local instance-level and global model-level feature attributions, while LIME constructs local surrogate linear models around individual predictions.

5Memorize the Three Data Acts (데이터 3법) Classifications: Know the legal boundaries between personal information (개인정보), pseudonymized information (가명정보, which can be processed without individual consent for statistical, scientific, and public archiving purposes), and fully anonymized information (익명정보, which is exempt from privacy statutes).

6Learn AI Cybersecurity Attack Vectors: Distinguish prompt injection (direct and indirect instruction manipulation in LLMs) from data poisoning (contaminating training sets to introduce backdoors) and model extraction/inversion (querying models to reconstruct proprietary weights or private client data).

Frequently Asked Questions

What is the KBI Financial AI Literacy (K-ALFA) examination?

The K-ALFA (KBI 금융 AI 리터러시) is an official professional qualification administered by the Korea Banking Institute (한국금융연수원 / KBI) designed to evaluate financial practitioners' knowledge of artificial intelligence fundamentals, business applications in banking and finance, and compliance with ethical and regulatory frameworks such as the FSC Financial AI Guidelines.

What is the examination format, duration, and passing score?

The official KBI sitting is a 90-minute, 100-point, four-option CBT. KBI does not publish an official item count on the K-ALFA qualification page checked 2026-09-20. Passing is 60 of 100 points.

What topics are tested on the K-ALFA examination?

The syllabus is structured around three primary domains: (1) AI Fundamentals and Financial Data Engineering (40%), (2) Financial AI Applications such as Robo-advisors, Alternative Credit Scoring, and FDS (25%), and (3) Financial AI Ethics, Regulation, and Governance, including FSC guidelines, XAI (SHAP/LIME), and AI security (35%).

What are South Korea's Financial AI Guidelines (금융분야 AI 가이드라인)?

Issued by the Financial Services Commission (FSC), the Financial AI Guidelines provide foundational principles for financial institutions developing and adopting AI systems. They mandate internal governance, data quality controls, algorithmic fairness, transparency and explainability for high-impact financial decisions, and cybersecurity defenses against model vulnerabilities.

How much does the K-ALFA exam cost and how often is it offered?

The examination fee is KRW 55,000. Testing sessions are organized periodically throughout the year at Korea Banking Institute testing facilities across South Korea.

Why does this practice question bank include Korean terminology alongside English?

The official examination uses standard South Korean financial and regulatory terms (such as '금융분야 AI 가이드라인', '이상거래탐지시스템 (FDS)', and '설명가능한 인공지능 (XAI)'). Providing both precise English explanations and official Korean technical terms ensures candidates achieve deep conceptual clarity while recognizing the exact terminology used on test day.