Bank Marketing — Analysis & Explanation

This document explains the operations performed on the bank marketing dataset and shows representative example outputs (tables and charts).

1) Load dataset — what this does

Reads the dataset into a tabular structure and presents a preview so you can verify columns, sample values, and basic shape.

Example preview (first 5 rows)

age	job	marital	education	balance	y
30	admin.	married	university.degree	1789	no
34	technician	single	high.school	0	no
47	blue-collar	married	basic.9y	1506	yes
22	services	single	high.school	0	no
58	retired	married	illiterate	214	no

2) Check for missing values & data types — what this does

Counts missing entries per column and reports each column's data type so you can plan cleaning and type conversions.

Example missing-values summary

column	missing
age	0
job	0
education	2
balance	0
y	0

Example data types

column	type
age	integer
job	categorical
balance	float
y	categorical (target)

3) Summary statistics — what this does

Computes descriptive statistics for numeric and categorical fields (count, mean, std, top categories, unique counts).

Example numeric summary (selected columns)

metric	age	balance
count	4521	4521
mean	41.7	1362.3
std	10.2	3045.1
min	18	-6847
25%	33	71
50%	39	448
75%	50	1428
max	95	102127

4) Target distribution — what this does

Shows counts of the target classes (e.g., how many subscribed vs not). Useful to detect class imbalance and to plan sampling strategies.

class	count	percent
no	3985	88%
yes	536	12%

5) Correlation matrix (numerical features) — what this does

Computes Pearson correlations between numeric fields to reveal linear relationships and potential multicollinearity.

Interpretation notes: high positive values (red) indicate strong positive correlation; negative values (blue) indicate inverse relationships. Use this info to decide feature selection or regularization.

6) Conclusion & recommended next steps

Address missing values in categorical columns (impute or mark as missing).
Consider balancing the target (undersample/oversample) before modeling.
Use correlation info to detect collinear features; drop or combine as needed.
Engineer features (e.g., bin ages, encode categories) and then train models with cross-validation.