Predictive Analytics in BI Cheat Sheet

Updated 2026-05-26

Next Topic: Pyramid Analytics Decision Intelligence Platform Cheat Sheet

🧠Study flashcards on this topic173 cards · spaced repetition→

Predictive analytics in business intelligence applies statistical algorithms, machine learning techniques, and historical data to forecast future outcomes and identify trends. It sits at the intersection of data science, BI tools, and business strategy, transforming raw data into actionable foresight that drives proactive decision-making. Unlike descriptive analytics that explains what happened, predictive analytics answers what will likely happen and why. The key mental model to remember: predictive analytics builds mathematical representations of reality (models) that capture patterns from the past to project into the future—but model quality depends entirely on data quality, feature selection, and validation rigor.

What This Cheat Sheet Covers

This topic spans 21 focused tables and 175 indexed concepts, 173 flashcards. Below is a complete table-by-table outline of this topic, spanning foundational concepts through advanced details.

Table 1: Regression TechniquesTable 2: Time Series Forecasting MethodsTable 3: Classification AlgorithmsTable 4: Ensemble MethodsTable 5: Model Validation TechniquesTable 6: Regression Accuracy MetricsTable 7: Classification MetricsTable 8: Feature Engineering TechniquesTable 9: Feature Selection MethodsTable 10: Correlation and Statistical AnalysisTable 11: Trend Analysis TechniquesTable 12: Anomaly and Outlier DetectionTable 13: Confidence and Prediction IntervalsTable 14: What-If and Scenario AnalysisTable 15: Hyperparameter Tuning MethodsTable 16: Business Applications of Predictive AnalyticsTable 17: Data Quality and Preprocessing IssuesTable 18: Model Interpretability & ExplainabilityTable 19: Survival AnalysisTable 20: Causal Inference & Uplift ModelingTable 21: Model Monitoring & Drift Detection

Quick IndexSubscribe to unlock

A jump-to index of every table row in this cheat sheet.

Mind MapSubscribe to unlock

An interactive map of every table and concept in this topic.

Table 1: Regression Techniques

Regression is the backbone of continuous-outcome prediction; every BI analyst should know when to use plain linear models versus regularized or logistic variants. The right choice depends on the number of predictors, the presence of multicollinearity, and whether the outcome is continuous or binary.

Technique	Example	Description
Linear Regression	`y = β0 + β1x + ε` `sales = 50 + 3*ads + error`	Predicts continuous outcomes by modeling the linear relationship between dependent and independent variables using ordinary least squares.
Multiple Linear Regression	`y = β0 + β1x1 + β2x2 + ... + βnxn` `sales = 50 + 3ads + 2season`	• Extends simple regression to multiple predictors • each coefficient represents the effect of one variable holding others constant
Logistic Regression	$P(y=1) = \frac{1}{1+e^{-(\beta_0+\beta_1x)}}$ `churn_prob = logit(score)`	Predicts binary outcomes using a sigmoid function — outputs probabilities 0–1 for classification despite the "regression" name.
Ridge Regression (L2)	`Loss = MSE + λ∑β²` `alpha=1.0 in sklearn`	Adds L2 penalty to shrink coefficients toward zero — prevents overfitting without eliminating features.
Lasso Regression (L1)	`Loss = MSE + λ∑\|β\|` `alpha=0.1 in sklearn`	Uses L1 penalty to force some coefficients to exactly zero — performs automatic feature selection by eliminating irrelevant predictors.
Elastic Net	`Loss = MSE + λ1∑\|β\|` `+ λ2∑β²` `l1_ratio=0.5`	Combines Ridge and Lasso penalties — balances feature selection with shrinkage, ideal when predictors are correlated.
Polynomial Regression	`y = β0 + β1x + β2x² + β3x³` `revenue = 10 + 5time + 0.2time²`	Models nonlinear relationships by adding polynomial terms — captures curves but risks overfitting at high degrees.
Quantile Regression	`Q50(y\|x) = β0 + β1x` `quantreg(formula, tau=0.9)`	Estimates conditional quantiles (e.g., median or 90th percentile) of the response — robust to outliers and useful for asymmetric or skewed outcomes.

Table 2: Time Series Forecasting Methods

Time series forecasting is central to BI use cases from demand planning to financial projections. Methods range from classical statistical models (ARIMA, ETS) that require stationarity and explicit seasonality handling, to modern deep learning architectures that learn complex temporal patterns directly from data.

Method	Example	Description
ARIMA (AutoRegressive Integrated Moving Average)	`ARIMA(2,1,1)` `order=(p,d,q)`	Models autocorrelations in stationary data using past values (AR), differencing (I), and past errors (MA) — foundational classical method.

The rest of this section is part of the full cheat sheet

Table 3: Classification Algorithms

Classification assigns observations to discrete categories — churn or not, fraud or legitimate, high-value or low-value. Choosing the right algorithm depends on dataset size, interpretability requirements, and whether the classes are balanced.

Algorithm	Example	Description
Decision Tree	`if revenue > 50k then high` `else if age < 30 then low`	Creates hierarchical if-then rules by splitting data on feature thresholds — interpretable but prone to overfitting without pruning.

The rest of this section is part of the full cheat sheet

Table 4: Ensemble Methods

Ensemble methods combine multiple models to produce predictions more accurate and robust than any single model. Bagging reduces variance; boosting reduces bias; stacking leverages complementary model strengths.

Method	Example	Description
XGBoost (Extreme Gradient Boosting)	`XGBClassifier(max_depth=6)` `objective='binary:logistic'`	Highly optimized gradient boosting with regularization, parallel processing, and built-in cross-validation — industry standard for tabular data.

The rest of this section is part of the full cheat sheet

Table 1: Regression Techniques

Technique	Example	Description
Linear Regression	`y = β0 + β1x + ε` `sales = 50 + 3*ads + error`	Predicts continuous outcomes by modeling the linear relationship between dependent and independent variables using ordinary least squares.
Multiple Linear Regression	`y = β0 + β1x1 + β2x2 + ... + βnxn` `sales = 50 + 3ads + 2season`	• Extends simple regression to multiple predictors • each coefficient represents the effect of one variable holding others constant
Logistic Regression	$P(y=1) = \frac{1}{1+e^{-(\beta_0+\beta_1x)}}$ `churn_prob = logit(score)`	Predicts binary outcomes using a sigmoid function — outputs probabilities 0–1 for classification despite the "regression" name.
Ridge Regression (L2)	`Loss = MSE + λ∑β²` `alpha=1.0 in sklearn`	Adds L2 penalty to shrink coefficients toward zero — prevents overfitting without eliminating features.
Lasso Regression (L1)	`Loss = MSE + λ∑\|β\|` `alpha=0.1 in sklearn`	Uses L1 penalty to force some coefficients to exactly zero — performs automatic feature selection by eliminating irrelevant predictors.
Elastic Net	`Loss = MSE + λ1∑\|β\|` `+ λ2∑β²` `l1_ratio=0.5`	Combines Ridge and Lasso penalties — balances feature selection with shrinkage, ideal when predictors are correlated.
Polynomial Regression	`y = β0 + β1x + β2x² + β3x³` `revenue = 10 + 5time + 0.2time²`	Models nonlinear relationships by adding polynomial terms — captures curves but risks overfitting at high degrees.
Quantile Regression	`Q50(y\|x) = β0 + β1x` `quantreg(formula, tau=0.9)`	Estimates conditional quantiles (e.g., median or 90th percentile) of the response — robust to outliers and useful for asymmetric or skewed outcomes.

Table 2: Time Series Forecasting Methods

Method	Example	Description
ARIMA (AutoRegressive Integrated Moving Average)	`ARIMA(2,1,1)` `order=(p,d,q)`	Models autocorrelations in stationary data using past values (AR), differencing (I), and past errors (MA) — foundational classical method.

Table 3: Classification Algorithms

Algorithm	Example	Description
Decision Tree	`if revenue > 50k then high` `else if age < 30 then low`	Creates hierarchical if-then rules by splitting data on feature thresholds — interpretable but prone to overfitting without pruning.

Table 4: Ensemble Methods

Method	Example	Description
XGBoost (Extreme Gradient Boosting)	`XGBClassifier(max_depth=6)` `objective='binary:logistic'`	Highly optimized gradient boosting with regularization, parallel processing, and built-in cross-validation — industry standard for tabular data.