index Introduction
Concepts, Techniques and Tools

Dimension Reduction & Multivariate

How do we simplify complexity?
Set 02 · Level 02 of the Analytics Ladder

When You Have Too Many Variables — Simplify Without Losing Signal

The final frontier of the analytics ladder. Turn high-dimensional, complex data into structures humans can understand and act on.

02Analytics Ladder · Level 02 of 10

Modern data is high-dimensional. A customer survey has 40 questions. A financial model has 80 indicators. A genomics dataset has thousands of gene expression values. Dimensionality reduction compresses this complexity into a smaller number of meaningful dimensions without discarding the essential information.

These techniques serve two distinct purposes: computational (reducing variables before running another model, removing multicollinearity, speeding up algorithms) and interpretive (visualising complex data, discovering hidden structure, making patterns legible to the human eye).

You cannot understand 40 variables at once. Dimensionality reduction finds the 2 or 3 underlying directions that explain most of what matters — and lets you see the structure clearly.
7
Techniques
PC
Principal
Components
2D
Visualisation
from n dimensions
3
Tools Covered
Technique 01

Principal Component Analysis (PCA)

Transforms correlated variables into uncorrelated principal components ordered by the variance they explain. The workhorse of dimensionality reduction — linear, fast, interpretable.

Technique 02

Factor Analysis (FA)

Identifies latent factors underlying observed variables. Unlike PCA (variance-focused), FA models shared variance and is explicitly theory-driven. Covered here as a dimensionality reduction complement to PCA.

Technique 03

t-SNE

t-Distributed Stochastic Neighbour Embedding. Non-linear technique for 2D/3D visualisation. Preserves local structure — nearby points in high-dimensional space cluster together in the plot. Excellent for exploratory analysis and cluster discovery.

Technique 04

UMAP

Uniform Manifold Approximation and Projection. Faster than t-SNE, better at preserving global structure. Increasingly preferred for large-scale exploratory visualisation in genomics, customer analytics, and text analysis.

Technique 05

Canonical Correlation Analysis (CCA)

Examines the relationship between two sets of variables simultaneously. Finds linear combinations of each set that are maximally correlated. The multivariate extension of simple correlation.

Technique 06

Multidimensional Scaling (MDS)

Visualises the similarity or dissimilarity between objects in a low-dimensional map. Starts from a distance matrix (not raw data) and arranges items so that similar items are close together in 2D.

Technique 07

Autoencoders (intro)

Neural network architecture for non-linear dimensionality reduction. An encoder compresses the input to a low-dimensional "bottleneck"; a decoder reconstructs the original. The bottleneck representation is the reduced-dimension encoding.

PCA vs Factor Analysis — when to use each

PCA: Use when the goal is data compression, removing multicollinearity before regression, or preparing data for machine learning. No underlying theory required — just variance maximisation. Factor Analysis: Use when you have a theoretical model of latent constructs (e.g. "intelligence," "brand equity," "service quality") and want to test whether observed variables reflect those constructs. Theory-driven, confirmatory or exploratory.

Technique 01 · Part A

Principal Component Analysis — Foundations

PCA finds the directions of maximum variance in high-dimensional data and projects everything onto those directions.

Imagine 50 customer satisfaction questions, most of which are correlated with each other. PCA finds a small number of principal components — new variables that are linear combinations of the original questions — that capture most of the variation in the data. The first principal component captures the most variance; the second captures the most of what remains; and so on.

The Three Core Concepts

Concept 01

Variance Explained

Each principal component explains a percentage of the total variance in the data. The first component typically explains the most. A good PCA retains components that together explain 70–80% of total variance.

Reported as a "scree plot" (eigenvalue vs component number) and a "cumulative variance explained" table.

Concept 02

Loadings

Each component has a loading for each original variable — showing how strongly that variable contributes to the component. High positive loading = the variable rises when the component rises. High negative loading = inverse relationship.

Use loadings to name components: a component with high loadings on "queue time," "response speed," and "resolution time" is a "service speed" dimension.

Concept 03

Scores

Each observation (customer, product, period) receives a score on each component. Scores are the new reduced-dimension representation of the data — they replace the original 50 variables with 3–5 component scores.

Use scores in subsequent analysis: cluster analysis, regression, or visualisation — each customer now has 3 coordinates instead of 50.

Concept 04

Orthogonality

All principal components are uncorrelated with each other (orthogonal). This is a critical advantage: it eliminates multicollinearity when using PCA scores as predictors in regression.

This is the fundamental difference from the original variables, which were correlated. PCA transforms correlated inputs into uncorrelated components.

How many components to retain?

RuleCriterionHow to applyLimitation
Kaiser CriterionRetain components with eigenvalue > 1.0Each component should explain more variance than a single original variable (eigenvalue = 1.0 is the baseline)Can retain too many components with large datasets
Scree Plot ElbowRetain components before the "elbow" bendPlot eigenvalues; retain components before the line flattens sharplyElbow is sometimes ambiguous or gradual
Cumulative VarianceRetain enough components to explain 70–80% of varianceSum eigenvalues as percentages; stop when cumulative % exceeds thresholdThreshold is arbitrary — 70% may discard important detail
InterpretabilityRetain components that can be meaningfully namedA component whose loadings form no interpretable pattern is noise — drop itSubjective — but essential for business communication
Always standardise before PCA

PCA maximises variance. If variables are in different units (revenue in $000s vs satisfaction on 1–10), the high-variance variable (revenue) will dominate the first component entirely — regardless of its analytical importance. Always standardise all variables to mean=0, SD=1 (Z-scores) before running PCA. In SPSS: tick "Correlation matrix" in the Factor Analysis options. In Python: StandardScaler() before PCA().

Technique 01 · Part B

PCA — Reading & Interpreting Output

The maths runs itself. The analytical skill is in reading the output and translating it into business meaning.

Full Worked Example
Customer Experience Survey — 10 variables reduced to 3 components

A company surveys 500 customers on 10 aspects of their experience. PCA reveals 3 components explaining 74% of total variance:

Survey ItemPC1 LoadingPC2 LoadingPC3 Loading
Queue wait time−0.820.120.08
Resolution speed−0.790.080.15
First-call resolution−0.750.180.11
Staff friendliness0.140.810.09
Staff knowledge0.090.760.18
Problem understanding0.210.720.15
Website ease of use0.080.110.85
App responsiveness0.110.090.80
Digital self-service0.150.140.77
Overall NPS−0.410.440.38
% Variance31%26%17%

PC1: High negative loadings on wait time, resolution speed, first-call resolution → label: "Service Efficiency" (lower scores = faster, better service).

PC2: High positive loadings on friendliness, knowledge, problem understanding → label: "Staff Quality".

PC3: High positive loadings on website, app, digital self-service → label: "Digital Experience".

The 10-variable survey is now described by 3 actionable dimensions: Efficiency, Staff Quality, Digital. Each customer has a score on each — enabling targeted interventions by segment.

Using PCA scores in downstream analysis

Next stepHow PCA scores helpAdvantage over raw variables
RegressionUse PC scores as predictors. E.g. predict NPS from Efficiency, Staff Quality, Digital scores.Eliminates multicollinearity among predictors. Coefficients are stable and interpretable.
ClusteringRun K-means on 3 PC scores instead of 10 raw variables. Faster and avoids scale issues.Fewer dimensions means less noise; clustering is more stable.
VisualisationPlot PC1 vs PC2 as a 2D scatter plot. Each customer is a point — spatial proximity = similar experience.Makes 10-dimensional data visually legible. Cluster structure and outliers become visible.
Anomaly detectionObservations with extreme scores on any component are outliers relative to the component's dimension.Detects unusual customers, fraudulent patterns, or measurement errors efficiently.
Technique 02

Factor Analysis as Dimensionality Reduction

While PCA extracts variance, Factor Analysis extracts meaning — the latent constructs that explain why variables correlate.

Set 05 introduced EFA and CFA in the context of testing theoretical models. Here we revisit Factor Analysis specifically as a dimensionality reduction tool — its role in compressing many observed variables into a smaller number of interpretable constructs before further analysis.

PCA vs Factor Analysis — a clear comparison

DimensionPCAFactor Analysis
GoalMaximise variance explained in the dataExplain shared variance — the co-variation due to latent factors
Theory needed?No — purely data-drivenOptional for EFA; required for CFA
Components/FactorsOrdered by variance explained (PC1 captures most)Factors represent underlying constructs — order is not inherent
UniquenessAll variance is accounted for by componentsEach variable has a unique variance component (error) not attributed to any factor
RotationOptional (orthogonal/oblique)Standard — Varimax (orthogonal) or Oblimin (oblique) improves interpretability
Output used forData compression, feature engineering for ML, removing multicollinearityScale construction, survey validation, theoretical construct testing
Primary toolSPSS, Python (sklearn.decomposition.PCA)SPSS, IBM AMOS, Python (factor_analyzer)
When to choose Factor Analysis over PCA for dimensionality reduction

Choose FA when: (1) Your variables are survey items or scale measures designed to reflect underlying constructs. (2) You need to validate construct validity (Cronbach's alpha, AVE, CR). (3) You plan to use the factors in SEM (Set 05). (4) You want rotation to improve interpretability — FA rotations are theoretically motivated, PCA rotations are not. Choose PCA when you simply need to reduce dimensionality without a theoretical model.

Business Application
Employee Engagement Scale — From 20 Items to 4 Factors

An HR team administers a 20-item engagement survey. EFA with Oblimin rotation identifies 4 factors: Autonomy (items 1–5), Relationship Quality (items 6–10), Growth Opportunities (items 11–15), Recognition (items 16–20). Each factor's items are averaged to create a factor score. These 4 scores replace the 20 raw items in subsequent regression predicting turnover intention. Result: a clean, interpretable model where each factor's unique contribution to turnover can be estimated — with no multicollinearity.

Techniques 03 & 04

t-SNE & UMAP

Non-linear dimensionality reduction for visualisation. See the structure in high-dimensional data that PCA cannot reveal.

PCA is linear — it only finds straight-line directions of maximum variance. Many real datasets have complex, curved, non-linear structures that PCA misses entirely. t-SNE and UMAP are non-linear techniques specifically designed to produce 2D or 3D visualisations where meaningful clusters and structures in high-dimensional data become visible.

t-SNE — t-Distributed Stochastic Neighbour Embedding

t-SNE converts high-dimensional distances between points into probabilities, then finds a 2D arrangement of points that best preserves those probability relationships. Points that were nearby in high-dimensional space end up clustered together in the 2D plot.

PropertyDetail
PerplexityThe most important parameter. Controls the "neighbourhood size" — how many nearby points t-SNE considers when calculating probabilities. Typical range: 5–50. Lower = focuses on very local structure. Higher = considers broader neighbourhood. Try multiple values and compare.
Local vs Global structuret-SNE preserves local structure (nearby points cluster correctly) but does NOT preserve global structure (the distance between clusters on the plot is not meaningful). Never interpret the relative distances between clusters in a t-SNE plot.
StochasticResults vary between runs (random initialisation). Set a random seed for reproducibility. Run multiple times to confirm that cluster patterns are stable.
ScaleSlow on large datasets (>50,000 points). Use approximate t-SNE (BH-t-SNE) or switch to UMAP for large data.

UMAP — Uniform Manifold Approximation and Projection

UMAP is a newer technique that addresses t-SNE's limitations while producing equally compelling visualisations. It is faster, scales better to large datasets, preserves more global structure, and can be used as a general-purpose dimensionality reduction tool (not just for visualisation).

UMAP advantage 01

Speed

UMAP runs 10–100× faster than t-SNE on large datasets. Practical for exploratory analysis of millions of records in minutes.

UMAP advantage 02

Global structure preserved

Unlike t-SNE, the relative positions of clusters in a UMAP plot are more meaningful — similar clusters in high-dimensional space appear closer in the 2D plot.

UMAP advantage 03

Deterministic option

UMAP can be set to produce the same result every run, making it easier to incorporate into reproducible analysis pipelines.

Key parameter

n_neighbors

Controls the balance between local and global structure. Small values (5–15) emphasise fine local detail. Larger values (30–100) emphasise broad global patterns.

⚠ Critical limitations of t-SNE and UMAP

Both are for exploration only. The 2D plots cannot be used for quantitative analysis, prediction, or statistical testing. They show that clusters exist — not why. The axes have no meaningful units. Cluster sizes in the plot are artefacts of the algorithm, not reflections of actual group sizes. Use these for hypothesis generation and visual communication — not for drawing analytical conclusions.

Business Application
Customer Behaviour Visualisation — 50 variables in 2D

A telecom company has 80,000 customers described by 50 behavioural variables (call patterns, data usage, tenure, payment history, etc.). PCA reduces to 10 components (explaining 72% of variance). UMAP then projects those 10 components to 2D. The 2D plot reveals 5 distinct customer clusters — which are then profiled (Set 07 methods) and used to design targeted retention programmes. The entire pipeline: standardise → PCA → UMAP → cluster → profile → act.

Techniques 05 · 06 · 07

CCA, MDS & Autoencoders

Three more advanced techniques for specific multivariate reduction problems.

Canonical Correlation Analysis (CCA)

CCA finds the relationship between two sets of variables simultaneously — the multivariate extension of simple correlation. It identifies linear combinations of Set 1 variables and Set 2 variables that are maximally correlated with each other.

Business Example
CCA — Employee experience and customer outcomes

Set 1 (Employee): Autonomy score, Manager rating, Workload satisfaction, Growth perception (4 variables).
Set 2 (Customer): NPS, Satisfaction score, Repeat purchase rate, Complaint rate (4 variables).

CCA finds the combination of employee experience variables that best predicts the combination of customer outcome variables. First canonical variate: Autonomy + Manager rating → NPS + Satisfaction (r = 0.68). Interpretation: employees who feel autonomous and well-managed produce customers who score higher on NPS and satisfaction simultaneously.

Multidimensional Scaling (MDS)

MDS takes a distance or dissimilarity matrix (not raw data) and produces a 2D or 3D map where items that are most similar are placed closest together. It reveals the perceptual or conceptual structure underlying a set of pairwise comparisons.

Business Example
MDS — Brand Perception Map

Customers rate the similarity of 8 competing brands on a 1–10 scale (how similar is Brand A to Brand B?). MDS converts these similarity ratings into a 2D perceptual map. The two axes that emerge (after interpretation) represent: X-axis = "Value vs Premium" · Y-axis = "Traditional vs Innovative." Brands cluster into quadrants: premium-traditional (Brands A, B), value-innovative (Brands E, F), etc. The map shows where white space exists for product positioning and reveals which brands are direct competitors.

Autoencoders — Non-linear Dimensionality Reduction

An autoencoder is a neural network trained to compress input data into a lower-dimensional "bottleneck" layer and then reconstruct the original data from that compressed representation. The bottleneck activations are the reduced-dimension features.

ComponentRoleBusiness meaning
EncoderCompresses input (e.g. 100 variables) to bottleneck (e.g. 5 dimensions)Learns the most efficient compressed representation of the data
Bottleneck layerThe reduced-dimension representationThese 5 values replace the original 100 — the learned features
DecoderReconstructs the original data from the bottleneckReconstruction quality validates the encoding — poor reconstruction = information lost
Reconstruction errorDifference between input and decoded outputHigh error for a specific observation = anomaly (used in fraud detection)
Autoencoders vs PCA

PCA is linear — it can only find straight-line compressions. Autoencoders are non-linear (due to activation functions) — they can find complex curved compressions that better capture non-linear structure. Rule of thumb: try PCA first (faster, interpretable). If PCA leaves too much unexplained, try an autoencoder. In Python: tensorflow.keras or torch.nn. For anomaly detection specifically, autoencoders are the preferred approach — observations with high reconstruction error are flagged as anomalous.

Tool Guide 🟡 Go Deeper

SPSS & Excel

PCA and Factor Analysis in SPSS. Manual PCA walkthrough in Excel.

🟡 Go DeeperSPSS — PCA & Factor Analysis

SPSS runs PCA and Factor Analysis through the same menu — the distinction is made in the extraction method settings.

01
Open Factor Analysis
Analyze → Dimension Reduction → Factor
02
Add variables
Move all your numeric variables to the Variables box. These should already be standardised — or tick "Correlation matrix" in Extraction to automatically standardise.
03
Set extraction method
Click Extraction:
• For PCA: Method = Principal Components
• For FA: Method = Principal Axis Factoring or Maximum Likelihood
Analyse: Correlation matrix (this standardises variables automatically)
Extract: Eigenvalues over 1 (Kaiser criterion) or specify number of factors
Tick Unrotated factor solution and Scree plot → Continue
04
Set rotation
Click Rotation:
• Orthogonal (uncorrelated factors): Varimax
• Oblique (correlated factors, more realistic): Direct Oblimin
Tick Rotated solution and Loading plot → Continue
05
Request scores and options
Click Scores → tick Save as variables (saves component scores as new dataset columns) → Regression method → Continue
Click Options → Suppress small coefficients: 0.40 → Continue → OK
Key SPSS output to read
KMO and Bartlett's test (suitability check — KMO >0.60) · Communalities table · Total Variance Explained table (eigenvalues, % variance, cumulative %) · Scree plot · Rotated Component Matrix (loadings per variable per component)
06
Name and use the components
Review the Rotated Component Matrix. For each component, identify the variables with loadings ≥ 0.40. Name each component based on the pattern of high-loading variables. The saved factor score variables (FAC1_1, FAC2_1, etc.) can now be used in regression, clustering, or further analysis.
🟢 Start Here (Manual PCA)Excel — Manual PCA with Data Analysis ToolPak

Excel does not have a built-in PCA function, but the Data Analysis ToolPak provides the components needed to do it manually — useful for understanding the mechanics.

01
Standardise all variables
For each column: create a new column = =(X - AVERAGE($X$2:$X$n)) / STDEV($X$2:$X$n). Or use Analyze → Descriptive Statistics → Save Z-scores if copying from SPSS output.
02
Generate correlation matrix
Data → Data Analysis → Correlation → select standardised columns → OK. The correlation matrix is the basis for PCA.
03
Plot cumulative variance
If you have eigenvalues from SPSS or Python, paste them into Excel and calculate cumulative % with =SUM($B$2:B2)/SUM($B$2:$B$n)*100. Insert a line chart to create the scree/cumulative variance plot for presentation.
04
Compute component scores manually
If loadings (from SPSS/Python) are in columns, compute PC scores with =SUMPRODUCT(standardised_row_values, loading_column). Drag down for all observations. Paste these scores as values before further analysis.
Tool Guide 🔵 Full Power

Python — sklearn, umap-learn & more

Complete dimensionality reduction pipeline — from PCA through t-SNE, UMAP, and autoencoder basics.

🔵 Full PowerPCA — Full Pipeline
# pip install scikit-learn matplotlib seaborn import pandas as pd import numpy as np import matplotlib.pyplot as plt import seaborn as sns from sklearn.preprocessing import StandardScaler from sklearn.decomposition import PCA # 1. Standardise scaler = StandardScaler() X_std = scaler.fit_transform(df.select_dtypes(include='number')) # 2. Fit PCA (keep all components first to assess) pca_full = PCA() pca_full.fit(X_std) # 3. Scree plot + cumulative variance explained = pca_full.explained_variance_ratio_ * 100 cumulative = np.cumsum(explained) fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(12, 4)) ax1.bar(range(1, len(explained)+1), explained, color='#1a6b6b', alpha=0.8) ax1.set_title('Scree Plot — % Variance per Component') ax1.set_xlabel('Principal Component') ax1.set_ylabel('% Variance Explained') ax2.plot(cumulative, marker='o', color='#1a6b6b') ax2.axhline(80, color='orange', linestyle='--', label='80% threshold') ax2.set_title('Cumulative Variance Explained') ax2.legend() plt.tight_layout() plt.show() # 4. Retain n components (e.g. 3) n_components = 3 pca = PCA(n_components=n_components) scores = pca.fit_transform(X_std) # 5. Loadings table loadings = pd.DataFrame( pca.components_.T, index=df.select_dtypes(include='number').columns, columns=[f'PC{i+1}' for i in range(n_components)] ) print(loadings.round(3)) # 6. Biplot (PC1 vs PC2 with variable arrows) scores_df = pd.DataFrame(scores, columns=[f'PC{i+1}' for i in range(n_components)]) plt.figure(figsize=(9, 7)) plt.scatter(scores_df['PC1'], scores_df['PC2'], alpha=0.4, color='#2c5f8a') for i, var in enumerate(loadings.index): plt.arrow(0, 0, loadings.loc[var,'PC1']*3, loadings.loc[var,'PC2']*3, head_width=0.05, color='#c0392b', alpha=0.8) plt.text(loadings.loc[var,'PC1']*3.2, loadings.loc[var,'PC2']*3.2, var, fontsize=9, color='#c0392b') plt.title('PCA Biplot — PC1 vs PC2') plt.axhline(0, color='grey', linewidth=0.5) plt.axvline(0, color='grey', linewidth=0.5) plt.show()
🔵 Full Powert-SNE & UMAP Visualisation
# t-SNE from sklearn.manifold import TSNE tsne = TSNE(n_components=2, perplexity=30, random_state=42, n_iter=1000) tsne_coords = tsne.fit_transform(X_std) plt.figure(figsize=(9, 7)) plt.scatter(tsne_coords[:,0], tsne_coords[:,1], c=df['cluster_label'], cmap='tab10', alpha=0.6, s=20) plt.colorbar(label='Cluster') plt.title(f't-SNE Visualisation (perplexity=30)') plt.xlabel('t-SNE 1 (no units)') plt.ylabel('t-SNE 2 (no units)') plt.show() # UMAP — pip install umap-learn import umap reducer = umap.UMAP(n_neighbors=30, min_dist=0.1, random_state=42) umap_coords = reducer.fit_transform(X_std) plt.figure(figsize=(9, 7)) plt.scatter(umap_coords[:,0], umap_coords[:,1], c=df['cluster_label'], cmap='tab10', alpha=0.6, s=20) plt.title('UMAP Visualisation') plt.show() # Common pipeline: PCA → t-SNE (PCA pre-reduces noise before t-SNE) pca_50 = PCA(n_components=50, random_state=42) X_pca50 = pca_50.fit_transform(X_std) tsne2 = TSNE(n_components=2, perplexity=30, random_state=42) coords = tsne2.fit_transform(X_pca50) # faster, less noisy
🔵 Full PowerAutoencoder for Anomaly Detection
# Simple autoencoder — pip install tensorflow from tensorflow import keras from tensorflow.keras import layers n_features = X_std.shape[1] encoding_dim = 5 # bottleneck: 5 dimensions from n_features # Build autoencoder input_layer = keras.Input(shape=(n_features,)) encoded = layers.Dense(32, activation='relu')(input_layer) encoded = layers.Dense(encoding_dim, activation='relu')(encoded) decoded = layers.Dense(32, activation='relu')(encoded) decoded = layers.Dense(n_features, activation='linear')(decoded) autoencoder = keras.Model(input_layer, decoded) autoencoder.compile(optimizer='adam', loss='mse') autoencoder.fit(X_std, X_std, epochs=50, batch_size=32, validation_split=0.1) # Anomaly detection: high reconstruction error = anomaly reconstructed = autoencoder.predict(X_std) recon_error = np.mean((X_std - reconstructed)**2, axis=1) threshold = np.percentile(recon_error, 95) # top 5% flagged df['anomaly'] = recon_error > threshold print(f"Anomalies detected: {df['anomaly'].sum()}")
Interactive Practice

Three Simulators

Build intuition for PCA, component interpretation, and the effect of dimensionality on data structure.

Simulator 1 · PCA Variance Explorer

A 10-variable dataset is analysed by PCA. The eigenvalues are shown. Adjust the cumulative variance threshold and see how many components you need to retain — then evaluate the trade-off.

Adjust the threshold to see how many components are retained at each level of explained variance.

Simulator 2 · PCA Loading Pattern Interpreter

Three PCA loading tables are shown. For each, interpret the loading pattern and name the component based on which variables load highly.

Component Pattern 1 of 3
Review the loadings and select the most appropriate name for this component.

Simulator 3 · Technique Selector

Six analytical situations are described. Match each to the correct dimensionality reduction technique.

For each scenario, select the most appropriate technique from the dropdown.
Best Practice

Common Mistakes

Dimensionality reduction mistakes are particularly hard to detect — the output always looks plausible.

Mistake 01

Not standardising before PCA

A variable with large numerical values will dominate the first component regardless of its analytical importance. Annual revenue in $000s will overwhelm a satisfaction scale (1–10). Always Z-score all variables first. In SPSS: use Correlation Matrix extraction. In Python: StandardScaler() before PCA().

Mistake 02

Interpreting t-SNE/UMAP inter-cluster distances

The distance between clusters in a t-SNE or UMAP plot is not meaningful. Two clusters that appear far apart may not be more dissimilar than two clusters appearing close. These plots reveal cluster existence and membership — not quantitative inter-cluster relationships.

Mistake 03

Retaining too few components without checking

Applying Kaiser's rule rigidly (eigenvalue > 1) may keep too few or too many components. Always cross-check with the scree plot and with interpretability. A component that cannot be meaningfully named is worth questioning — even if its eigenvalue exceeds 1.

Mistake 04

Confusing PCA components with factor analysis factors

Components (PCA) are mathematical constructions — they explain variance and may span all original variables. Factors (FA) are theoretical constructs — they model shared variance and each variable should load cleanly on one factor. The same label is sometimes misapplied to both, but they have different assumptions, outputs, and uses.

Mistake 05

Using PCA without validating that the data is suitable

Run the KMO (Kaiser-Meyer-Olkin) test and Bartlett's test of sphericity before PCA/FA. KMO < 0.6 suggests the variables are not sufficiently correlated for factor analysis — PCA would extract mostly noise. Bartlett's p > 0.05 means the correlation matrix is essentially an identity matrix — no structure to extract.

Mistake 06

Applying t-SNE to small datasets or for quantitative analysis

t-SNE needs at minimum several hundred observations to produce stable, meaningful layouts. On small datasets (n < 200), results are unreliable. More importantly: t-SNE output cannot be used for statistical testing, regression, or any quantitative analysis. It is a visualisation tool — not an analytical one.

Set 10 · Key Takeaway
Dimensionality reduction reveals the structure hidden inside complexity. But the structure you find is only as meaningful as the care taken in preparation and interpretation.
Final Assessment

Knowledge Check

Five questions across the techniques of Set 10 — and the final completion of the 10-Set Series.

Q1 — A PCA on 12 customer survey variables produces eigenvalues: 3.8, 2.4, 1.6, 0.9, 0.7, (remaining <0.5). Using the Kaiser criterion and the scree plot elbow, how many components should be retained?
A
5 components — retain all with eigenvalue close to 1.0
B
3 components — eigenvalues 3.8, 2.4, and 1.6 all exceed 1.0 (Kaiser criterion), and the elbow in the scree plot occurs between components 3 and 4 (0.9 is a sharp drop from 1.6)
C
2 components — only those with eigenvalue above 2.0
D
All 12 — retain all components to avoid losing information
Kaiser criterion: retain eigenvalues >1.0 → Components 1 (3.8), 2 (2.4), and 3 (1.6) qualify. Component 4 (0.9) fails. The scree plot elbow: there is a sharp drop from 1.6 to 0.9 between components 3 and 4 — the "elbow" occurs there. Both rules agree: retain 3 components. Cumulative variance = (3.8+2.4+1.6)/12 = 65% — acceptable. If 65% is insufficient for the use case, try 4 components and assess interpretability.
Q2 — A PCA loading table shows Component 1 has high positive loadings on "Queue time" (0.81), "Hold duration" (0.78), and "Response speed" (0.75). What should this component be named?
A
Component 1 — always use the component number as the label
B
Customer Satisfaction — all variables relate to service
C
Service Speed (or Process Efficiency) — all three high-loading variables measure time-related aspects of how quickly service is delivered
D
Cannot be named — PCA components have no meaning
Naming components is the critical interpretive step that turns mathematical output into business insight. All three high-loading variables (queue time, hold duration, response speed) measure speed-related service delivery — not quality, friendliness, or outcome. "Service Speed" or "Process Efficiency" accurately captures the shared theme. A component that cannot be meaningfully named from its loadings may be extracting noise — consider dropping it.
Q3 — A t-SNE plot of 10,000 customer records (50 variables) shows 4 distinct clusters. A colleague interprets the large physical distance between Cluster 1 and Cluster 4 as evidence that these are "very different" customers. What is wrong with this interpretation?
A
Nothing — inter-cluster distance on a t-SNE plot is the most important output
B
t-SNE does not preserve inter-cluster distances meaningfully. The physical distance between clusters in a t-SNE plot is an artefact of the algorithm — it does not reflect actual dissimilarity between clusters. Use profiling (Set 07) and statistical tests to quantify how different the clusters actually are.
C
t-SNE should have used perplexity=100 to produce accurate distances
D
Only 2 clusters are visible so the 4-cluster interpretation is wrong
t-SNE optimises to preserve local neighbourhood structure — it does NOT preserve the quantitative distances between distant groups. Two clusters that appear far apart may or may not be more dissimilar than two clusters appearing close. t-SNE output correctly tells you: "these clusters exist" and "these observations belong together." It does not tell you "how different" the clusters are from each other. To quantify inter-cluster differences: run ANOVA or discriminant analysis on the cluster labels using the original variables.
Q4 — An analyst wants to remove multicollinearity from 15 financial indicators before running a regression model. Which technique is most appropriate?
A
PCA — it transforms the correlated indicators into uncorrelated principal components, which can then be used as regression predictors without multicollinearity
B
t-SNE — visualise the data first to confirm the variables are correlated
C
Factor Analysis — run EFA to find the latent financial constructs
D
UMAP — reduce to 2D before regression
PCA produces orthogonal (uncorrelated) components — this is its specific advantage for solving multicollinearity in regression. Run PCA on the 15 indicators → retain 3–5 components explaining 75–80% of variance → use those component scores as predictors. Components are mathematically guaranteed to be uncorrelated, eliminating multicollinearity entirely. t-SNE and UMAP are for visualisation, not feature engineering. Factor Analysis could work but assumes latent theoretical constructs — less appropriate than PCA for pure data compression before regression.
Q5 — A KMO test returns 0.54 before running PCA on a set of 20 variables. What does this indicate?
A
The data is excellent for PCA — KMO above 0.5 is ideal
B
The data is marginal for PCA — KMO below 0.60 suggests the variables are not sufficiently correlated. Review the variable selection: remove variables with low communalities and re-test before proceeding.
C
KMO is irrelevant — proceed with PCA regardless
D
Run more observations to improve KMO above 0.5
KMO measures sampling adequacy — specifically whether the variables share enough common variance (correlation) to justify factor/component extraction. KMO = 0.54 is in the "miserable" range (0.5–0.6). The variables are not strongly enough intercorrelated for meaningful PCA/FA. Actions: (1) Check anti-image correlation matrix for variables with low individual KMO scores — remove the weakest ones. (2) Re-examine whether all 20 variables are conceptually related. (3) Re-run KMO after removal. Do not proceed with PCA until KMO exceeds 0.60, ideally 0.70+.

🎓 Advanced Analytical Skills — Series Complete

You have completed all 10 sets of the Advanced Analytical Skills series. From Set 01's descriptive analysis to Set 10's dimensionality reduction, you have covered the full spectrum of the analytics ladder — the techniques, tools, business applications, and practical judgement that characterise a highly capable analyst.

01
Descriptive
02
Diagnostic
03
Forecasting
04
Drivers
05
Multivariate
06
Classification
07
Segmentation
08
Risk
09
Optimisation
10
Dimensionality

The analytical skill is not in knowing which button to press — it is in knowing which question to ask, which technique is appropriate, and how to translate the output into decisions that create value.