Download This the business Intellignce notes. and more Lecture notes Business Statistics in PDF only on Docsity!
Unit II
Modelling in Business Intelligence
Contents:
Models and modelling in BI, Model Presentation, Model Building, Model Assessment and
Quality of Models, Modelling using Logical Structures: ontology & Frame, Modelling using
graph structure: Business process model and notation (BPMN), Modelling using probabilistic
structures, Modelling using analytical structure. Model and Data: data Generation, The Role
of time, Data Quality.
Models and modelling in BI:
Model Presentation:
Model Presentation refers to the process of communicating analytical model results to decision-makers in a clear, meaningful, and actionable way. In Business Intelligence (BI), models may generate complex outputs such as forecasts, risk scores, optimization results, or probability estimates. However, if these outputs are not properly interpreted and presented, decision-makers may struggle to understand or trust them. Therefore, model presentation bridges the gap between technical analysis and business strategy.
An effective model presentation translates technical results into business language. Instead of showing raw statistical values, it explains what the results mean for sales growth, cost reduction, risk exposure, or operational efficiency. The focus is on answering key business questions such as: What is happening? Why is it happening? What will happen next? What action should be taken?
Several presentation tools and formats are commonly used in BI systems. Interactive dashboards allow managers to monitor real-time KPIs and explore data through filters and drill-down features. KPI scorecards compare actual performance against targets. Heat maps and trend charts help visualize patterns, correlations, and anomalies. Scenario comparison charts allow decision-makers to evaluate multiple alternatives side by side. Executive summaries provide concise insights for senior management without overwhelming them with technical complexity.
Important principles:
● Clarity: Visuals should be simple and easy to interpret. Avoid unnecessary complexity.
● Relevance: Present only information that supports decision-making. ● Accuracy: Ensure that visual representations correctly reflect model outputs. ● Timeliness: Information should be up-to-date and delivered when needed. ● Consistency: Use uniform metrics and formats across reports to avoid confusion.
Another critical aspect is data storytelling. This involves presenting results in a logical narrative structure — starting from the problem, explaining the analysis, highlighting insights, and concluding with recommendations. Storytelling improves engagement and ensures that decision-makers clearly understand the implications.
Modern BI tools such as Microsoft Power BI, Tableau, and SAP BusinessObjects provide advanced visualization and reporting capabilities. These tools allow dynamic interaction, predictive insights, and automated reporting, making model presentation more powerful and user-friendly.
Model Building:
Model Building is the structured process of developing a mathematical, statistical, or logical representation of a real-world business problem. In Business Intelligence (BI), models are built to analyze patterns, predict future outcomes, optimize processes, or support strategic decisions. A well-built model simplifies complex business situations while maintaining accuracy and reliability.
The first step in model building is problem definition. Clearly defining the objective is crucial because it determines what type of model is required. For example, a company may want to forecast sales, detect fraud, optimize inventory, or predict customer churn. Without a clear objective, the model may produce irrelevant or misleading results.
Next is data collection and preparation. Relevant data is gathered from databases, transaction systems, customer records, or external sources. Since raw data often contains missing values, errors, or inconsistencies, data cleaning and preprocessing are necessary to improve data quality. This step may include normalization, handling missing values, removing outliers, and transforming categorical variables.
After cleaning, feature selection is performed. Features (variables) are the factors that influence the outcome. Selecting the right variables improves model performance and reduces complexity. For example, in sales forecasting, important features may include price, seasonality, advertising spend, customer demographics, and economic conditions.
The next step is choosing the appropriate modelling technique. The choice depends on the problem type:
● Statistical models (e.g., regression) for prediction. ● Mathematical optimization models for resource allocation. ● Machine learning models for pattern recognition. ● Logical models for rule-based decisions.
Common tools and libraries used in model building include Python (with libraries like Scikit-learn), R, and platforms like SAS. These tools help in implementing algorithms, training models, and evaluating performance.
Once the model type is selected, training and tuning are performed. Training involves feeding historical data into the model so it can learn patterns. Tuning adjusts parameters to improve accuracy and avoid problems such as overfitting (where the model performs well on training data but poorly on new data).
Finally, the model is validated and tested using separate datasets to measure performance. Metrics such as accuracy, precision, recall, mean squared error, or R-squared are used depending on the problem type.
“Order contains Product,” and “Supplier provides Product.” By clearly defining these relationships, ontology ensures consistency in reporting and analytics.
Ontology improves:
● Data integration across multiple systems ● Semantic consistency in BI reports ● Interoperability between software applications ● Knowledge sharing within organizations
In advanced BI systems, ontology supports semantic querying and intelligent search. Technologies such as Protégé and standards like OWL are commonly used for building ontologies.
Frame-Based Modelling: Frame-based modelling represents knowledge using structured templates called frames. Each frame describes an entity and includes attributes (slots) and their values. Frames resemble object-oriented structures where each object has properties and characteristics.
For example, a “Customer Frame” may include:
● Name ● Age ● Contact Details ● Purchase History ● Loyalty Status
Similarly, a “Product Frame” may include:
● Product ID ● Category ● Price ● Stock Level
Frames allow hierarchical relationships, where one frame can inherit properties from another. For instance, a “Premium Customer” frame may inherit attributes from the “Customer” frame but include additional benefits or privileges.
Frame-based models are particularly useful for:
● Rule-based systems ● Expert systems ● AI-driven recommendation engines ● Semantic Business Intelligence
Importance in BI: Logical modelling using ontology and frames enhances knowledge representation and supports intelligent decision-making. It moves BI beyond traditional numerical analysis and enables semantic understanding of data. This approach is especially valuable in AI-enabled BI platforms, where systems need to understand relationships, meanings, and rules rather than just raw numbers.
Modelling using graph structure:
Business process model and notation (BPMN):
Graph-based modelling represents business processes visually using nodes and connecting
flows. BPMN is a globally accepted standard for designing and documenting workflows in a
structured and easy-to-understand format. It uses standardized symbols such as events (start,
intermediate, end), activities/tasks , gateways (decision or branching points), sequence
flows , and message flows to describe how processes move from one step to another.
BPMN diagrams improve communication between business managers and technical teams by providing a common visual language. For example, in an order processing system, BPMN
can show steps such as order placement, payment verification, inventory check, shipping, and
delivery. Decision gateways can represent scenarios like payment success or stock
availability.
Advanced BPMN models may include:
● Swimlanes to represent responsibilities of departments ● Parallel flows to show simultaneous activities ● Exception handling paths for error management
By visualizing processes, organizations can identify bottlenecks, redundancies, delays, and compliance gaps. BPMN also supports automation by linking process models with workflow
management systems. Overall, it enhances operational efficiency, transparency, and process
optimization.
Modelling using Probabilistic Structures
Probabilistic modelling is used when business decisions involve uncertainty, risk, or incomplete information. These models apply probability theory to estimate the likelihood of
events and predict possible outcomes. Instead of giving exact answers, they provide
probabilities and risk assessments.
Common probabilistic models include:
● Bayesian Networks: Represent relationships between variables and update probabilities when new information is available. ● Markov Models: Predict future states based on current states, widely used in customer behavior and inventory forecasting. ● Monte Carlo Simulation: Uses repeated random sampling to estimate possible outcomes and risk levels.
In finance, probabilistic models assess credit risk and investment uncertainty. In marketing,
they predict customer churn probability. In healthcare, they estimate disease risk. These
● Batch data processing (collected periodically) ● Real-time data streaming (continuous updates) ● Big Data generation from high-volume platforms Modern systems such as SAP ERP and Salesforce automatically generate and store structured business data. Additionally, technologies like IoT devices and e-commerce platforms significantly increase data volume.
The Role of time:
● Dynamic Nature of Data: Business data continuously changes due to daily transactions, customer activities, and market conditions, making time an essential dimension in BI analysis.
● Trend Analysis: Time helps identify patterns and movements in data over days, months, or years, allowing businesses to understand growth or decline.
● Time-Series Forecasting: Historical time-based data is used to predict future outcomes such as sales demand, revenue, or customer behavior.
● Performance Comparison: Organizations compare current performance with past results (e.g., this month vs last month or this year vs last year) to measure improvement.
● Seasonality Analysis: Time enables detection of seasonal trends such as festival sales, holiday demand, or cyclical market patterns.
● Real-Time Monitoring: Time-stamped data supports real-time dashboards that monitor live business operations and allow quick responses to changes.
● Historical Analysis: Past data records help in identifying long-term growth patterns and evaluating business strategies.
● Accurate Database Management: Proper handling of timestamps, date formats, and time zones ensures reliable BI modelling and reporting.
Data Quality:
Data quality is essential in Business Intelligence because all analysis, reports, and dashboards depend on accurate and reliable data. If the input data is incorrect, the output insights will
also be incorrect. This follows the principle of “Garbage In, Garbage Out (GIGO),” meaning
poor data leads to poor decisions.
Accuracy: Data must correctly represent real-world values without errors.
Completeness: All required data fields should be filled, with no missing or blank records.
Consistency: Data formats and values should be uniform across different systems (e.g., same
currency or date format). Timeliness: Data must be up-to-date and available when needed for decision-making.
Validity: Data should follow defined business rules and standards.
Uniqueness: There should be no duplicate records in the database.
Methods to Improve Data Quality
- Data Cleansing and Validation: Removing errors, correcting incorrect entries, and verifying accuracy.
- Standardization Rules: Ensuring uniform data formats and consistent coding standards.
- ETL (Extract, Transform, Load) Processes: Cleaning and transforming data before loading it into data warehouses.
- Data Governance Policies: Establishing rules and responsibilities for managing data.
- Regular Audits and Monitoring: Continuously checking and updating data to maintain quality.
b) Identify the key components of Model Building in BI.
Ans:
Model building in Business Intelligence (BI) involves several structured components that ensure the model accurately represents a business problem and produces reliable insights. The key components are as follows:
1. Problem Definition The first and most important component is clearly defining the business problem or objective. This includes identifying what needs to be predicted, analyzed, or optimized. A clear objective ensures the model aligns with business goals and avoids irrelevant analysis. 2. Data Collection Relevant data is gathered from internal systems (sales, finance, CRM, ERP) and external sources (market data, social media, third-party datasets). The quality and relevance of collected data directly affect model performance. 3. Data Preparation and Cleaning Raw data often contains missing values, duplicates, or inconsistencies. This component involves cleaning, transforming, and organizing data to make it suitable for analysis. Techniques include handling missing values, removing outliers, and formatting data correctly. 4. Feature Selection Feature selection involves identifying the most important variables that influence the outcome. Selecting relevant features improves model accuracy and reduces complexity. Irrelevant variables are removed to avoid noise in the model. 5. Selection of Modelling Technique Choosing the appropriate modelling approach is essential. Depending on the problem, this may include statistical models (regression), machine learning models (classification, clustering), optimization models, or time-series forecasting techniques. 6. Model Training The selected model is trained using historical data. During training, the model learns patterns and relationships between input variables and the target outcome. 7. Model Tuning Model parameters are adjusted to improve performance and avoid issues such as overfitting or underfitting. Hyperparameter tuning helps optimize accuracy and reliability. 8. Model Validation and Testing The model is tested using separate datasets to evaluate its performance. Metrics such as accuracy, precision, recall, or error rates are used to measure effectiveness. 9. Deployment and Monitoring After validation, the model is implemented in the BI system for real-time or regular use. Continuous monitoring is required to ensure performance remains consistent as business conditions change.
c) Utilize Business Process Model and Notation (BPMN) to model a
complex business process in a given case study.
Ans:
Case Study: Online Retail Order Fulfillment Process Consider an e-commerce company that manages customer orders through multiple departments such as Sales, Finance, Warehouse, and Logistics. The process includes order placement, payment verification, inventory check, shipping, and delivery confirmation. This
is a complex process because it involves multiple decision points, parallel activities, and exception handling. BPMN Model Description Using BPMN, the process can be structured as follows:
1. Start Event The process begins with a Start Event when a customer places an online order. 2. Order Verification Task The Sales system records the order details and forwards the request for payment verification. 3. Gateway – Payment Approval Decision A Gateway (Decision Point) checks whether the payment is successful: ● If Payment Successful → Proceed to inventory check.
● If Payment Failed → Notify customer and end process.
4. Inventory Check Task The Warehouse system checks product availability. 5. Gateway – Stock Availability Decision ● If Stock Available → Proceed to packaging and shipping.
● If Stock Not Available → Notify customer and initiate refund process.
6. Parallel Tasks (Packaging & Invoice Generation) Using a Parallel Gateway , two activities occur simultaneously: ● Package the product.
● Generate invoice and update system records.
7. Shipping Task The packaged product is handed over to the Logistics department for delivery. 8. Delivery Confirmation Event Once delivered, the system updates order status and sends confirmation to the customer. 9. End Event The process ends after successful delivery and confirmation.
Benefits of Using BPMN in This Case ● Provides a visual representation of the entire workflow
● Identifies bottlenecks (e.g., delays in payment approval)
● Highlights decision points and exception handling
● Improves coordination between departments
● Supports process automation and optimization
b) Describe the factors that contribute to the Quality of Models in Business
Intelligence.
Ans:
The quality of models in Business Intelligence (BI) determines how accurately and reliably they support decision-making. A high-quality model provides meaningful insights, performs consistently, and aligns with business objectives. Several important factors contribute to ensuring model quality. A well-defined business objective is the foundation of a quality model. If the problem is unclear or poorly framed, the model may produce irrelevant or misleading results. The model must align with specific business goals such as forecasting sales, detecting fraud, or optimizing inventory.
1. High-Quality Data Data quality is one of the most critical factors. Accurate, complete, consistent, and timely data ensures reliable model outputs. Poor-quality data leads to incorrect predictions and weak model performance. Data preprocessing and validation play a key role in maintaining quality. 2. Appropriate Model Selection Choosing the correct modelling technique based on the problem type improves effectiveness. For example, regression models are suitable for prediction, classification models for categorization, and optimization models for resource allocation. Incorrect model selection reduces accuracy and usefulness. 3. Proper Feature Selection Selecting relevant variables (features) significantly improves model performance. Including unnecessary or irrelevant features may introduce noise and reduce accuracy. Proper feature engineering enhances predictive power. 4. Accuracy and Performance Metrics Model quality is evaluated using performance metrics such as accuracy, precision, recall, F1-score, Mean Squared Error (MSE), or R-squared values. These metrics help measure how well the model performs on unseen data. 5. Avoidance of Overfitting and Underfitting A quality model must balance complexity. Overfitting occurs when the model learns noise instead of patterns, while underfitting happens when it fails to capture important relationships. Techniques like cross-validation and regularization help maintain balance. 6. Robustness and Generalization The model should perform well with new or changing data. Robust models are stable and adaptable to variations in business conditions. 7. Interpretability and Transparency Decision-makers must understand how the model produces results. Transparent models build trust and are especially important in finance, healthcare, and compliance-related decisions. 8. Continuous Monitoring and Updating Business environments change over time. Regular monitoring, recalibration, and retraining ensure that the model remains accurate and relevant.
c) Develop a Model Presentation for a hypothetical business scenario and
justify your design choices.
Ans:
Hypothetical Scenario: Retail Sales Performance and Demand Forecasting A retail company wants to analyze its sales performance and forecast future demand to improve inventory management and maximize profit. A predictive model has been developed
to forecast monthly sales based on historical sales data, promotions, seasonality, and customer trends. The next step is to present the model results effectively to management. Proposed Model Presentation Design
1. Executive Summary Dashboard The presentation begins with a high-level dashboard created using a BI tool such as Microsoft Power BI. This dashboard shows key KPIs such as: ● Total Monthly Sales ● Forecasted Sales for Next 3 Months ● Percentage Growth Rate ● Inventory Turnover Ratio Justification: Senior managers prefer concise and visual summaries. A KPI-focused dashboard quickly communicates performance without overwhelming them with technical details. 2. Trend Analysis Chart A time-series line chart displays historical sales alongside forecasted sales for upcoming months. Justification: A line chart clearly shows patterns, seasonality, and projected growth trends. Comparing past and predicted values improves confidence in forecasting results. 3. Category-Wise Sales Breakdown A bar chart presents sales performance by product category (e.g., Electronics, Clothing, Groceries). Justification: Category comparison helps management identify high-performing and low-performing segments, supporting strategic decisions like promotional focus or stock adjustments. 4. Scenario Comparison Panel A scenario analysis section compares three possible situations: ● Normal demand ● High-demand (festival season) ● Low-demand (off-season) Justification: Scenario comparison supports risk assessment and contingency planning, helping managers prepare for uncertainty. 5. Inventory Risk Heat Map A heat map highlights products at risk of overstocking or stockouts. Justification: Heat maps visually emphasize critical issues using color intensity, making it easy to detect problem areas quickly. 6. Drill-Down Capability Interactive filters allow users to view data by region, store location, or time period. Justification: Drill-down features provide flexibility, allowing managers to explore detailed insights without cluttering the main dashboard. Design Principles Applied ● Clarity: Simple visuals with limited text for easy understanding. ● Relevance: Only business-critical metrics are displayed. ● Visual Hierarchy: Important KPIs placed at the top. ● Interactivity: Enables deeper exploration when needed. ● Action-Oriented Insights: Recommendations included below charts
