Unit 8.5
Predictive Analytics: Forecasting What Will Happen
IT 233: Business Information Systems
Learning Objectives
By the end of this chapter, you will be able to:
- ✅ Define Predictive Analytics and the key question it answers.
- ✅ Describe main techniques like regression and classification.
- ✅ Understand the role of machine learning in making predictions.
- ✅ Provide business examples like fraud detection and demand forecasting.
Moving Beyond the Past
Predictive Analytics: The branch of advanced analytics that uses historical data, statistical algorithms, and machine learning to identify the likelihood of future outcomes.
Descriptive Analytics
"What happened?"
Summarizes past data.
⚡ Predictive Analytics
"What will happen?"
Forecasts future events.
🎬 Watch: What Predictive Analytics Is
The Core Question: "What will happen?"
Predictive analytics aims to answer forward-looking questions:
- Which customers are most likely to churn (leave)?
- What will our sales be for the next quarter?
- Is this credit card transaction likely to be fraudulent?
- What is the probability this marketing lead will convert?
The goal is to move from being reactive to proactive.
Key Techniques & Tools 📊
Predictive analytics uses a range of statistical and machine learning methods:
- Regression Analysis
- Classification Analysis
- Machine Learning (ML)
- Time-Series Forecasting
Let's explore each one...
Technique 1: Regression Analysis
Used to predict a continuous value (a number).
Example: Real Estate in Kathmandu
Question: What will be the selling price of a house?
Data Used:
- Area (in ana/ropani)
- Number of bedrooms
- Location (e.g., Baneshwor vs. Budhanilkantha)
Prediction: A specific price, like NPR 2.5 Crore.
Simulator: House Price Estimator
Adjust the features below to see how a regression model estimates a Kathmandu house price.
Formula: Base (NPR 15 lakh/ana) × area × location + bedrooms bonus
🧮 Worked Example: Reading a Regression Line
Illustrative: a clothing shop compares monthly Facebook ad spend with monthly sales (both in NPR thousand). The fitted line is:
- 50 = expected sales with zero ads (walk-in customers).
- 4 = each extra NPR 1,000 of ads is linked with about NPR 4,000 more sales.
- Prediction for NPR 20 thousand of ads: 50 + 4 × 20 = NPR 130 thousand.
- Caution: the line only holds within the range of past data. Spending 10× more won't give 10× the effect.
Technique 2: Classification Analysis
Used to predict a categorical outcome (a label or class).
Example: Bank Loan Application
Question: Will this applicant default on their loan?
Data Used:
- Applicant's income
- Credit history
- Loan amount
Prediction: A category, like "Default" or "No Default".
Simulator: Loan Default Classifier
Set the applicant profile and see the model's classification.
Advanced Techniques
🧠 Machine Learning (ML)
Algorithms are "trained" on historical data to learn patterns.
They then use these learned patterns to make predictions on new, unseen data.
Many modern predictive models are built on ML.
📈 Time-Series Forecasting
A specific model for predicting future values based on past time-ordered data.
Uses:
- Forecasting sales
- Stock prices (e.g., NEPSE index)
- Website traffic
🎓 How a Model Learns, and How It Cheats
- Training data (e.g. 80% of past records): the model learns patterns from it.
- Test data (the other 20%): kept hidden, then used to check the model on cases it hasn't seen.
- Overfitting: the model memorises the training data, noise and all, like a student who memorises past papers but fails new questions.
- Sign of overfitting: excellent on training data, poor on test data.
🎬 Watch: Machine Learning in 100 Seconds
Simulator: Time-Series Sales Forecast
Past 6 months of sales are shown. Click Show Forecast to reveal the model's 3-month prediction.
Practical Applications in Nepal 🎯
How local businesses can use predictive analytics:
Customer Churn Prediction
Ncell or Nepal Telecom identifying customers likely to switch providers.
Fraud Detection
eSewa or Khalti flagging suspicious transactions in real-time.
Demand Forecasting
Bhat-Bhateni Supermarket predicting demand for items during Dashain.
Credit Scoring
Banks assessing loan risk for individuals and businesses.
Match: Business Problem to Technique
Click a scenario, then click the matching technique. Green = correct, Red = wrong.
SCENARIOS
TECHNIQUES
🧪 Quick Activity: Regression, Classification or Time Series?
- How many passengers will ride a Sajha bus route next month, based on the last 3 years of monthly data? → Time series
- Will this student pass or fail the final exam? → Classification
- What room rate can a Pokhara hotel charge, given season, room size and lake view? → Regression
- Is this email spam or not spam? → Classification
🧮 Worked Example: A 99%-Accurate Useless Model
Illustrative: 1,000 card transactions, of which 10 are fraud.
Correct on 990 of 1,000 = 99% accurate.
Frauds caught: 0.
Catches 8 of the 10 frauds; 22 flags are false alarms.
Correct on 8 + 968 = 976 = 97.6% accurate, but far more useful.
Judge a model by the business cost of each mistake: a missed fraud vs. an annoyed customer.
A Word of Caution 🔍
This is crucial to remember:
Probabilities, Not Certainties
Predictive analytics provides an educated guess about the future, but it is not a crystal ball.
The value is not in 100% accuracy.
The value lies in using these probabilities to make better, more informed decisions.
🎓 Real Case: When a Prediction Becomes a Grade
In 2020, COVID cancelled England's A-level exams. The regulator, Ofqual, used an algorithm to predict each student's grade.
- It adjusted teachers' estimates using each school's past results.
- Many students, especially at schools with weaker histories, were downgraded below their teachers' estimates.
- After protests, the government withdrew the algorithm in August 2020 and used teacher-assessed grades instead.
🎬 Watch: The 2020 A-Level Algorithm
⚖️ Predictive Analytics: Benefits vs Limitations
- Act early (stock up, call a customer, block a fraud).
- Use resources where risk is highest.
- Plan with evidence, not guesswork.
- Only as good as past data; shocks (a pandemic, a blockade) break patterns.
- Biased data gives biased predictions.
- Hard to explain some models to customers or regulators.
Quiz: Descriptive or Predictive?
Click the correct analytics type for each business question.
"What were our total sales last month?"
"Which customers are likely to churn next quarter?"
"How many products did we sell during Dashain 2024?"
"Is this new Khalti transaction likely fraudulent?"
💬 Discuss: Should a Model Decide?
- Which decisions would you trust a prediction to make alone? Which need a human?
- A bank's model rejects your loan. What should the bank have to tell you?
- What past data might be misleading in Nepal right now?
Summary: Key Takeaways
- Predictive analytics uses historical data to forecast future outcomes.
- It answers the critical business question: "What will happen?"
- Key techniques include regression (for numbers) and classification (for categories), often powered by Machine Learning.
- It enables businesses to be proactive in areas like fraud detection, demand forecasting, and customer retention.
- It provides probabilities to guide smarter decisions, not guarantees.
Exam Practice 📝
Short answer
- What is predictive analytics?
- What question does predictive analytics answer?
- Give two business examples of predictive analytics.
Exam Practice 📝 (continued)
Long answer
- Explain predictive analytics, the techniques it uses, and its applications in business.
- Discuss the benefits and limitations of predictive analytics, with an example from Nepal.
Revise the answers in the full note.
Image & Video Credits
Images
- Richard Huish College Exam Hall: Lewser, public domain, via Wikimedia Commons
- Regression scatter diagram: course author (original), CC BY 4.0
Videos (YouTube)
- Local Government Association, "What is predictive analytics?" (0–80 s)
- Fireship, "Machine Learning Explained in 100 Seconds" (35–96 s)
- Sky News, "Shock and anger as A-level algorithm strips pupils of their grades" (2020; 54–122 s)
Thank You
Any Questions?
Next Up: Unit 8.6 - Prescriptive Analytics: How to Guide Business Decisions