4. Data Processing

Extracts useful info from raw data.

Similar to industrial process: input (raw data) → process → output (insights).

Applications: office automation, ticketing, time tracking, resource planning, forecasting, optimization.

Better data analysis & presentation.

Reduces data to meaningful info.

Easier storage & distribution.

Simplified report creation.

Improved productivity & profits.

More accurate decision-making.

Batch Processing

Online Processing

Real-time Processing

Distributed Processing

Time-sharing Processing

Processes stored data in groups (batch jobs).

Efficient for repetitive tasks.

Examples: monthly billing, reports, inventory.

Interactive, system processes data as it becomes available.

Uses internet, data centers, cloud computing.

Immediate response to inputs/events.

Provides instant outputs.

Higher costs but fast results.

Example: banking transactions.

Operations divided across multiple servers.

High fault tolerance (workload reallocated if node fails).

Example: Hadoop Distributed File System (HDFS).

Multiple users share one CPU in time slots.

Appears as exclusive access.

Typical for mainframe/supercomputers.

OLTP

OLAP

Online Transaction Processing

Handle read/write of individual records at high rate.

Transactional databases → many users simultaneously.

Used for day-to-day operations.

Example: bank account balance updates.

Not suitable for large data scans.

Online Analytical Processing

Designed for large analytical queries.

Not efficient for single record searches.

Focus: business insights, data summarization, decision-making.

Used by knowledge workers & decision-makers.

Good for interactive analytics.

Extract, Transform, Load

Integration layer between multiple structured/unstructured data sources & data warehouse.

Steps:

• Extract data from sources.

• Transform (clean, standardize, format).

• Load into warehouse.

Purpose: serve specific business goals (e.g., CRM, web analytics, partner data).

ETL tools automate data flow & enable regular updates.

Step before data analysis.

Tasks:

• Handle missing values.

• Remove redundant, incomplete, duplicate, incorrect records.

Goal: create final clean dataset for analysis.

Time depends on:

• Processing device power.

• Complexity & amount of input data.

• Storage location (e.g., data lake, database).

• Desired final output.

Algorithms extract knowledge, uncover properties, predict outcomes.

Supervised learning → future predictions (labeled data).

Unsupervised learning → explore data, hidden patterns (unlabeled data).

Key factors: data size/type, desired insights, use of results.

Uses labeled data.

Goal: assign data points to classes.

Examples: spam detection, image classification, fraud detection.

Predicts continuous numerical values.

Fits model to data to find dependencies.

Examples: house prices, energy forecasting, customer lifetime value.

Groups unlabeled data based on similarity.

Finds hidden patterns/structures.

Examples: customer segmentation, document clustering, market segmentation.

Translate machine learning outcomes into actions → guides company decisions.

Must be presented in a user-friendly form (tables, audio, video, images).

Actions (automatic or human-based) are more valuable than just insights.

Model performance judged via:

• Evaluation metrics

• Key Performance Indicators (KPIs).

Enhances understanding via graphical representation.

Key skill for data scientists.

Purposes: exploration, error detection, communication of results.

Visualization prevents misleading conclusions from raw numbers.

Common visualization types: tables, histograms, scatter plots, geomaps, charts, heat maps.

Useful for many variables or mix of numerical & categorical attributes.

Tips:

• Sort/group rows by important column.

• Sort columns by importance & relation.

• Highlight important values (colors, fonts).

Shows frequency distribution with bars.

X-axis = bins (value ranges), Y-axis = frequency.

Best for continuous numeric variables.

Key aspects:

• Equal bin sizes recommended.

• Absolute vs. relative frequency.

• Experiment with bin numbers for best clarity.

Plot two variables → show correlation (linear/nonlinear, pos/neg).

Common in regression analysis.

Characteristics:

• Adjust dot size to dataset.

• Limited to 2D → use PCA or multiple plots for multivariate data.

Show data using spatial arrangements & color scales.

Geomap → geographic regions with color scales.

Cartogram → distorts regions to reflect a variable.

Periodic table = special type of conceptual map.

Show changes over time.

Line = trends, Area = filled version for quantitative comparison.

Can compare multiple variables (e.g., sales vs. expenses).

Use bar lengths for categorical data.

Easy comparison of values, trends, minima & maxima.

Show proportions (sum = 100%).

Each slice = category (with color & legend).

Often annotated with percentages.

Combine bar & line graphs.

Useful when variables differ significantly.

Highlights comparisons across datasets.

Extend scatter plot → add color & size as variables (2–4D).

Used for analyzing patterns & correlations.

Risk: hard to read with too much data.

Can include interactivity (hover/click for details).

Represent values via colors.

Uses:

• Geographic (activity zones, e.g., football field).

• Web analytics (user clicks & engagement).

Light = low correlation/activity, Dark = high.

Useful in feature selection (drop less relevant variables).

Exploratory → Excel, Jupyter, R, Mathematica.

Presentation → Python, R (libraries for tables & charts).

Interactive → Dashboards (Python, Tableau).

Processed data must be saved in understandable formats → enables communication, sharing & decision-making.

Data format = structure & organization of data (rules, syntax, conventions).

Ensures compatibility across systems, software & platforms.

Should be machine-readable (type, size, structure info).

Should be human-readable for analysis & visualization.

Should be popular/standardized → interoperability.

Clear & simple formats help identify redundant or correlated data → more accurate processing.

XLS

CSV

XML

JSON

Protobuf

Apache Parquet

Microsoft Excel format.

Stores data in tables (rows = records, columns = variables).

Cells can contain text, numbers, formulas.

Supports summarization & visualization.

Longstanding, widely used in organizations.

Plain text format, tabular structure.

Each line = data record, values separated by commas.

First row = optional header.

Simple, easy for both humans & machines.

Markup language, flexible & standard for communication.

Human & machine-readable.

Uses tags (not predefined, user-defined).

Must follow rules (opening/closing tags, schema/DTD for validation).

<student>

<name>Charlie</name>

<lastname>Wood</lastname>

<age>20</age>

</student>

Lightweight, text-based, easy for humans & machines.

Widely used in web apps & APIs.

Written as key-value pairs inside curly braces { }.

Arrays use square brackets [ ].

Google’s system for serializing structured data.

Language & platform neutral.

Similar to XML but smaller & faster.

Structure defined first, then source code auto-generated for read/write.

Used in Java, Python, C++.

Column-oriented format in Hadoop ecosystem.

Efficient for big data & datasets with many columns.

Reads only required columns → faster & reduces I/O.

Provides better compression for same-type data.

Can store nested structures & access fields individually.

Best for large-scale data processing.

Defines how information is stored for quick access & retrieval.

Increasingly complex due to large volumes & access patterns.

Shift from single server → distributed storage:

• Uses multiple servers.

• Provides redundancy & scalability.

• Requires coordination (store, retrieve, process).

Data Warehouse

Data Lake

Data Mart

Stores relational data from business apps/systems.

Provides consolidated historical data for analysis.

Kept separate from operational systems.

Well-defined structure (schema on write).

Based on OLAP architecture.

Used for:

• Regular reporting (daily/weekly).

• Visualizations.

• Business Intelligence (BI).

Stores relational + non-relational data (e.g., IoT, social media).

Schema on read → flexible, all kinds of data.

Supports:

• SQL queries.

• Real-time analytics.

• Machine learning & predictive analytics.

Challenges: cataloging & securing unstructured data.

Best deployed in cloud → security, elasticity, scalability.

Subset of a data warehouse.

Stores information specific to a group of users/department.

Solves specific organizational problems.