train_info.md

Below is the basic info on train.csv

Column	Non-Null Count	Dtype
last contact date	39211	object
age	39211	int64
job	38982	object
marital	39211	object
education	37744	object
default	39211	object
balance	39211	int64
housing	39211	object
loan	39211	object
contact	28875	object
duration	39211	int64
campaign	39211	int64
pdays	39211	int64
previous	39211	int64
poutcome	9760	object
target	39211	object

1. Sample size: The dataset contains 39,211 records.

2. Categorical variables:

   • Job: 11 unique categories, with "blue-collar" being the most common (7,776 occurrences).
   • Marital status: 3 categories, "married" is most common (22,691).
   • Education: 3 levels, "secondary" is most frequent (19,584).
   • Default: Binary, "no" is predominant (36,954).
   • Housing: Binary, "yes" is more common (21,657).
   • Loan: Binary, "no" is more frequent (31,820).
   • Contact: Binary, "cellular" is more common (25,030).
   • Poutcome: 3 categories, "failure" is most frequent (4,949).
   • Target: Binary, "no" is predominant (33,384).

3. Numerical variables:

   • Age: Ranges from 18 to 95, with a mean of 42.12 years.
   • Balance: Ranges from -8,019 to 102,127, with a mean of 5,441.78. The large standard deviation (16,365.29) indicates high variability.
   • Duration: Ranges from 0 to 4,918, with a mean of 439.06 seconds (about 7.3 minutes).
   • Campaign: Ranges from 1 to 63, with a mean of 5.11 contacts.
   • Pdays: Ranges from -1 to 871, with a mean of 72.26. The -1 likely indicates no previous contact.
   • Previous: Ranges from 0 to 275, with a mean of 11.83 previous contacts.

4. Date variable:

   • Last contact date: 1,013 unique dates, with 2009-05-15 being the most frequent (313 occurrences).

Key observations:

1. The target variable is imbalanced, with "no" being much more common.
2. There's a wide range in account balances, with some negative balances.
3. The duration of calls varies greatly, from very short to over an hour.
4. Most customers haven't been contacted recently (median pdays is -1).
5. The number of campaign contacts varies widely, but is generally low (median is 2).
6. There's a significant portion of missing data in some variables (e.g., education, contact).

These insights align with the feature importance plot you shared earlier, highlighting why variables like duration, balance, and age are important predictors. The summary also reveals potential challenges like class imbalance and missing data that you'll need to address in your analysis or modeling.

Column	Missing Values
last contact date	0
age	0
job	229
marital	0
education	1467
default	0
balance	0
housing	0
loan	0
contact	10336
duration	0
campaign	0
pdays	0
previous	0
poutcome	29451
target	0