Customer Spending Forecasting
Juan Salazar
2024-04-15
Dataset Customer_data.csv
This dataset was grabbed from:
https://www.kaggle.com/datasets/sanyamgoyal401/customer-purchases-behaviour-dataset/data
id age gender income education region loyalty_status purchase_frequency
1 1 27 Male 40682 Bachelor East Gold frequent
2 2 29 Male 15317 Masters West Regular rare
3 3 37 Male 38849 Bachelor West Silver rare
4 4 30 Male 11568 HighSchool South Regular frequent
5 5 31 Female 46952 College North Regular occasional
6 6 38 Male 7347 Bachelor South Silver occasional
purchase_amount product_category promotion_usage satisfaction_score
1 18249 Books 0 6
2 4557 Clothing 1 6
3 11822 Clothing 0 6
4 4098 Food 0 7
5 19685 Clothing 1 5
6 2822 Electronics 0 5Model Summary
Call:
lm(formula = purchase_amount ~ satisfaction_score + age + income +
education_numeric + region_numeric + loyalty_numeric + purchase_frequency_numeric,
data = train_data)
Residuals:
Min 1Q Median 3Q Max
-9564.2 -774.3 0.3 768.2 9130.2
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) -1.032e+01 5.761e+01 -0.179 0.858
satisfaction_score 2.620e+00 5.533e+00 0.473 0.636
age 9.575e-01 1.283e+00 0.746 0.456
income 3.502e-01 4.426e-04 791.219 <2e-16 ***
education_numeric -1.823e+00 4.526e+00 -0.403 0.687
region_numeric -2.013e+00 5.230e+00 -0.385 0.700
loyalty_numeric -7.310e+00 9.605e+00 -0.761 0.447
purchase_frequency_numeric -3.032e+00 7.362e+00 -0.412 0.680
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Residual standard error: 1521 on 69994 degrees of freedom
Multiple R-squared: 0.8994, Adjusted R-squared: 0.8994
F-statistic: 8.944e+04 on 7 and 69994 DF, p-value: < 2.2e-16ResidualvsFitted Plot and Q_Q Residual Plot
Transforming Model
Estimated transformation parameter
Y1
0.6195993 Fitting our model
Call:
lm(formula = purchase_amount^0.5 ~ satisfaction_score + age +
income + education_numeric + region_numeric + loyalty_numeric +
purchase_frequency_numeric + product_category_numeric, data = train_data)
Residuals:
Min 1Q Median 3Q Max
-48.540 -5.061 0.208 5.298 36.182
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 4.248e+01 3.001e-01 141.565 <2e-16 ***
satisfaction_score 1.647e-02 2.826e-02 0.583 0.560
age 5.853e-03 6.554e-03 0.893 0.372
income 1.898e-03 2.260e-06 839.731 <2e-16 ***
education_numeric -3.158e-02 2.311e-02 -1.366 0.172
region_numeric -2.451e-03 2.671e-02 -0.092 0.927
loyalty_numeric -4.567e-02 4.905e-02 -0.931 0.352
purchase_frequency_numeric -2.979e-03 3.760e-02 -0.079 0.937
product_category_numeric -1.993e-02 1.654e-02 -1.205 0.228
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Residual standard error: 7.77 on 69993 degrees of freedom
Multiple R-squared: 0.9097, Adjusted R-squared: 0.9097
F-statistic: 8.815e+04 on 8 and 69993 DF, p-value: < 2.2e-16ResidualvsFitted Plot and Q_Q Residual Plot
Fixing Heteroscedasticity
Model: \(\text{Purchase Amount}^{0.5} = \beta_0 + \beta_1 \cdot \text{Satisfaction Score} + \beta_2 \cdot \text{Age} + \beta_3 \cdot \text{Income} + \beta_4 \cdot \text{Education} + \beta_5 \cdot \text{Region} + \beta_6 \cdot \text{Loyalty Status} + \beta_7 \cdot \text{Purchase Frequency} + \varepsilon\)
Fitted Model: \(\text{Purchase Amount}^{0.5} = \beta_0 + \beta_1 \cdot \text{Satisfaction Score} + \beta_2 \cdot \text{Age} + \beta_3 \cdot \text{Income} + \beta_4 \cdot \text{Education} + \beta_5 \cdot \text{Region} + \beta_6 \cdot \text{Loyalty Status} + \beta_7 \cdot \text{Purchase Frequency}\)
Call:
lm(formula = purchase_amount^0.5 ~ satisfaction_score + age +
income + education_numeric + region_numeric + loyalty_numeric +
purchase_frequency_numeric + product_category_numeric, data = train_data,
weights = 1/(fitted(lm_TRANS)^2))
Weighted Residuals:
Min 1Q Median 3Q Max
-0.38585 -0.05514 0.00238 0.05736 0.31041
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 3.926e+01 2.685e-01 146.224 <2e-16 ***
satisfaction_score -5.144e-03 2.552e-02 -0.202 0.840
age 4.120e-03 5.924e-03 0.695 0.487
income 2.028e-03 2.170e-06 934.670 <2e-16 ***
education_numeric -4.094e-02 2.089e-02 -1.960 0.050 *
region_numeric 5.875e-03 2.414e-02 0.243 0.808
loyalty_numeric -3.311e-02 4.422e-02 -0.749 0.454
purchase_frequency_numeric 1.371e-02 3.391e-02 0.404 0.686
product_category_numeric -2.100e-02 1.490e-02 -1.409 0.159
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Residual standard error: 0.08299 on 69993 degrees of freedom
Multiple R-squared: 0.9258, Adjusted R-squared: 0.9258
F-statistic: 1.092e+05 on 8 and 69993 DF, p-value: < 2.2e-16ResidualvsFitted Plot and Q_Q Residual Plot
Model Selection
Model: \(\text{Purchase Amount}^{0.5} = 39.21 + 0.00203 \cdot \text{Income} - 0.03754 \cdot \text{Education} + 0.05591 \cdot \text{Purchase Frequency} - 0.0333 \cdot \text{Product Category} + \varepsilon\)
Fitted Model: \(\text{Purchase Amount}^{0.5} = 39.21 + 0.00203 \cdot \text{Income} - 0.03754 \cdot \text{Education} + 0.05591 \cdot \text{Purchase Frequency} - 0.0333 \cdot \text{Product Category}\)
Start: AIC=-348474.3
purchase_amount^0.5 ~ satisfaction_score + age + income + education_numeric +
region_numeric + loyalty_numeric + purchase_frequency_numeric +
product_category_numeric
Df Sum of Sq RSS AIC
- satisfaction_score 1 0.0 482.0 -348476
- region_numeric 1 0.0 482.0 -348476
- purchase_frequency_numeric 1 0.0 482.0 -348476
- age 1 0.0 482.0 -348476
- loyalty_numeric 1 0.0 482.0 -348476
- product_category_numeric 1 0.0 482.0 -348474
<none> 482.0 -348474
- education_numeric 1 0.0 482.0 -348472
- income 1 6016.1 6498.1 -166380
Step: AIC=-348476.2
purchase_amount^0.5 ~ age + income + education_numeric + region_numeric +
loyalty_numeric + purchase_frequency_numeric + product_category_numeric
Df Sum of Sq RSS AIC
- region_numeric 1 0.0 482.0 -348478
- purchase_frequency_numeric 1 0.0 482.0 -348478
- age 1 0.0 482.0 -348478
- loyalty_numeric 1 0.0 482.0 -348478
- product_category_numeric 1 0.0 482.0 -348476
<none> 482.0 -348476
- education_numeric 1 0.0 482.0 -348474
- income 1 6016.3 6498.3 -166380
Step: AIC=-348478.2
purchase_amount^0.5 ~ age + income + education_numeric + loyalty_numeric +
purchase_frequency_numeric + product_category_numeric
Df Sum of Sq RSS AIC
- purchase_frequency_numeric 1 0.0 482.0 -348480
- age 1 0.0 482.0 -348480
- loyalty_numeric 1 0.0 482.0 -348480
- product_category_numeric 1 0.0 482.0 -348478
<none> 482.0 -348478
- education_numeric 1 0.0 482.0 -348476
- income 1 6016.5 6498.5 -166380
Step: AIC=-348480
purchase_amount^0.5 ~ age + income + education_numeric + loyalty_numeric +
product_category_numeric
Df Sum of Sq RSS AIC
- age 1 0.0 482.0 -348482
- loyalty_numeric 1 0.0 482.0 -348481
- product_category_numeric 1 0.0 482.0 -348480
<none> 482.0 -348480
- education_numeric 1 0.0 482.0 -348478
- income 1 6016.5 6498.5 -166381
Step: AIC=-348481.5
purchase_amount^0.5 ~ income + education_numeric + loyalty_numeric +
product_category_numeric
Df Sum of Sq RSS AIC
- loyalty_numeric 1 0.0 482.0 -348483
- product_category_numeric 1 0.0 482.0 -348482
<none> 482.0 -348482
- education_numeric 1 0.0 482.0 -348480
- income 1 6016.5 6498.5 -166383
Step: AIC=-348483
purchase_amount^0.5 ~ income + education_numeric + product_category_numeric
Df Sum of Sq RSS AIC
- product_category_numeric 1 0.0 482.0 -348483
<none> 482.0 -348483
- education_numeric 1 0.0 482.0 -348481
- income 1 6016.5 6498.5 -166385
Step: AIC=-348483
purchase_amount^0.5 ~ income + education_numeric
Df Sum of Sq RSS AIC
<none> 482.0 -348483
- education_numeric 1 0.0 482.1 -348481
- income 1 6016.6 6498.6 -166387
Call:
lm(formula = purchase_amount^0.5 ~ income + education_numeric,
data = train_data, weights = 1/(fitted(lm_TRANS)^2))
Weighted Residuals:
Min 1Q Median 3Q Max
-0.38562 -0.05511 0.00239 0.05736 0.31076
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 3.925e+01 7.659e-02 512.457 <2e-16 ***
income 2.028e-03 2.169e-06 934.722 <2e-16 ***
education_numeric -4.108e-02 2.089e-02 -1.967 0.0492 *
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Residual standard error: 0.08298 on 69999 degrees of freedom
Multiple R-squared: 0.9258, Adjusted R-squared: 0.9258
F-statistic: 4.369e+05 on 2 and 69999 DF, p-value: < 2.2e-16Plot Model Selection
Prediction vs Actual
id purchase_amount predicted_purchase_amount
1 1 18249 14811.544
2 2 4557 4931.719
3 3 11822 13920.628
7 7 3293 3132.324
10 10 8779 7946.495
13 13 2450 2781.652