INFORMATION TECHNOLOGY & DIGITAL BANKING MODULE B QUIZ 2

Operational systems are optimised for current transactions, while data warehouses are designed for integrated historical analysis and reporting. Analytical workloads can be resource-intensive, so separating them from transaction processing can protect operational performance.

A traditional data warehouse is subject-oriented, integrated, time-variant and non-volatile. It organises data around analytical subjects, reconciles data from multiple sources, preserves historical context and is mainly queried rather than continuously updated transaction by transaction.

ETL means Extract, Transform and Load: data are obtained from source systems, cleaned or converted into consistent formats, and loaded into the target warehouse. Some modern architectures also use ELT, where transformation occurs after loading into the target platform.

A staging area temporarily holds extracted source data for validation, cleansing, transformation and reconciliation before loading into curated warehouse structures. It can isolate processing from production systems and improve data-quality control.

A data mart is a subject- or department-focused subset of analytical data. It may be dependent on an enterprise warehouse or independently sourced, but governance is needed to prevent inconsistent definitions across marts.

OLAP supports multidimensional analysis, aggregation, trend analysis and complex queries across substantial historical datasets. OLTP focuses on high-volume, short, concurrent transactions such as posting a deposit or updating an account balance.

In a common layered architecture, source systems feed extraction/staging and integration processes, then curated warehouse or data-mart structures support reporting and analytics. Implementations vary, but analytical access is typically provided over prepared, governed data.

Data cleansing and entity resolution identify and correct inconsistent, duplicate or invalid records. Standardisation, matching rules, master data management and data-quality monitoring help maintain a consistent view across source systems.

Metadata describes data‘s meaning, structure, origin, transformations, ownership and usage. It helps analysts interpret measures consistently, trace lineage and investigate quality issues. Without it, similarly named fields may be misunderstood or compared incorrectly.

Data mining applies statistical, machine-learning or computational techniques to discover patterns, relationships, anomalies or predictive signals in data. Results should be validated and interpreted in context; a pattern is not automatically proof of fraud.

Classification is supervised learning because the training examples have known class labels. Clustering is generally unsupervised and groups observations by similarity without predefined outcome labels.

Association-rule mining identifies items or events that tend to occur together, often measured with support, confidence and lift. Such associations can inform cross-selling but do not by themselves establish causation or guarantee that an offer is suitable.

Regression models estimate numeric outcomes, such as a loss amount or demand level. Classification predicts categories, while the specific model choice should reflect assumptions, data quality, validation and the cost of prediction errors.

An expert system uses a knowledge base and an inference mechanism to apply rules or representations of specialist knowledge to a problem. Its output depends on the quality and completeness of the encoded knowledge and should be governed appropriately.

An ANN is a computational model made of interconnected units whose parameters are adjusted during training to learn patterns from data. It can model complex relationships, but may require substantial data and careful validation and may be less interpretable than simpler models.

Business intelligence (BI) turns data into information for monitoring and decision-making through reports, dashboards, visualisation and analysis. Reliable BI depends on clear metric definitions, data lineage and appropriate access controls.

A data lake can store large volumes of raw or lightly processed data in diverse formats, including structured, semi-structured and unstructured data. Governance, cataloguing, access control and data quality are still essential; a lake is not automatically well-organised or analytics-ready.

A scheduled batch refresh extracts and loads changes at defined intervals, often overnight. It can be cost-effective for periodic reporting but means the warehouse may not reflect the latest transaction immediately. Near-real-time or streaming pipelines are alternatives when freshness requirements justify them.

Stream processing evaluates continuously arriving data with low latency, enabling near-real-time detection and response. It requires suitable event pipelines, monitoring, quality controls and a plan for late, duplicated or out-of-order events.

Responsible analytics requires lawful and purpose-appropriate data use, minimisation, access controls, quality checks, model validation, monitoring for bias and clear accountability. Data should not be used simply because it is technically available.