Skip to content

sanasz91mdev/ml-spending-behaviour

Repository files navigation

Predicting Spending Interest Report

[[TOC]]

Data Summary:

Data Value
Problem Type Multiclass Classification
Number of rows on training data 2000
Training - to - Test Ratio 80 to 20
Feature count 15
Number of categorical features 5
Number of Numeric features 10
Target Variable Target-MCC1
Target Variable Type Categorical
Number of classes in Target 10
Values of classes in Target 4111, 4814, 5411, 5621, 5691, 5812, 5983,7032, 5999, 7832

MCCs Lookup - Definition

Merchant category code Description
7832 Motion Picture Theaters
5411 Grocery Stores
5812 Eating places and Restaurant
7032 Sporting and Recreational Camps
5999 Miscellaneous and Specialty Retail Stores
4111 Transportation
5691 Men’s and Women’s Clothing Stores
5983 Fuel Oil, Liquefied Petroleum
4814 Telecommunication services
5621 Women’s Ready-to-Wear Stores

Sample Top 5 Rows

AgeGroup Marital Status Day Gender City Avg-7832 Avg-5411 Avg-5812 Avg-7032 Avg-5983 Avg-4111 Avg-5999 Avg-5691 Avg-4814 Avg-5621 Target-MCC1
20 Single Weekday F KHI 10 25 20 10 5 5 5 5 15 0 5411
20 Single Weekday F KHI 0 10 0 65 25 0 0 0 0 0 7032
30 Single Weekday M KHI 10 5 0 0 32 0 15 0 0 38 5621
40 Single Weekday F ISL 0 22 10 3 3 5 7 2 0 48 5621
30 Married Weekday F ISL 10 25 20 10 5 5 5 5 15 0 5411

Cardinality:

Refers to the number of possible values that a categorical feature can assume:

Feature Number of Unique Values Unique Values
AgeGroup 3 20s, 30s, 40s
City 3 KHI, LHR, ISL
Day 2 Weekday, Weekend
Gender 2 Male, Female

ML Algorithm

Algorithm Value
Problem Type Multiclass Classification
Name Random Forest
Hyperparameter Values
n_estimators 100
random_state 42
max_depth none

Model Accuracy

Metric Value
Model Accuracy 0.93

Feature Importance

Feature Importance
AgeGroup 0.007335352846323242
MaritalStatus 0.003780795258948365
Day 0.006009442644614128
Gender 0.004859433183447916
City 0.006545369156478366
Avg-7832 0.46691761668599463
Avg-5411 0.03854568023937666
Avg-5812 0.01321235913167839
Avg-7032 0.2247785109812169
Avg-5983 0.01376515795764316
Avg-4111 0.10336592425084284
Avg-5999 0.03103362660596684
Avg-5691 0.016577535325028935
Avg-4814 0.013617798502985055
Avg-5621 0.04965539722945451

Classification Report:

          precision    recall  f1-score   support

    4111       0.71      0.55      0.62        22
    4814       1.00      1.00      1.00         1
    5411       0.92      0.95      0.93        58
    5621       0.98      1.00      0.99        89
    5691       0.88      0.95      0.91        56
    5812       0.92      0.85      0.88        40
    5983       1.00      1.00      1.00         7
    7032       0.98      0.96      0.97        55
    7832       0.93      0.94      0.94        72

accuracy                           0.93       400
macro avg       0.92      0.91     0.92       400
weighted avg    0.93      0.93     0.93       400

Definitions

Precision

It is the proportion of true positive predictions out of all the positive predictions. It measures the accuracy of positive predictions. It Answers the following question: what proportion of predicted Positives is truly Positive?

Precision = TP/(TP+FP)

Recall:

It is the proportion of true positive predictions out of all the actual positive cases. It measures the completeness of positive predictions. what proportion of actual Positives is correctly classified? recall is

Recall = TP/(TP+FN)

Accuracy:

what proportion of photos — both Positive and Negative — were correctly classified? proportion of correct predictions out of all the predictions made by the model on the test data accuracy is

Accuracy = (TP+TN)/(TP+TN+FP+FN)

F1-score:

It is the harmonic mean of precision and recall. It balances both the metrics and gives an overall measure of the model's performance.

F1-score = 2 × (precision × recall)/(precision + recall)

Support:

It is the number of actual occurrences of the class in the dataset.

The report includes these metrics for each class in the target variable. In this case, there are ten classes represented by the numbers 4111, 4814, 5411, 5621, 5691, 5812, 5983, 7032, 5999, 7832.

  • The macro average and weighted average of precision, recall, and F1-score are also provided.
  • Macro average is the unweighted average of each metric across all the classes.
  • Weighted average considers the number of samples in each class.

Releases

No releases published

Packages

No packages published