This project predicts on-time delivery of electronic products for an international e-commerce company, and studies delivery factors and customer behavior.
The dataset (10,999 observations, 12 variables) includes customer ID, warehouse block, shipment mode, customer care calls, customer rating, product cost, prior purchases, product importance, gender, discount offered, product weight, and delivery status.
Product weight and cost significantly affect delivery. Products weighing 2500 - 3500 grams and costing less than $250 are more likely to be delivered on time. Most products are shipped from Warehouse F via ship.
Customer behavior impacts delivery. More customer calls correlate with delayed delivery. Customers with more prior purchases have more on-time deliveries, indicating loyalty. Products with a discount of 0-10% are more likely to be late, while those with discounts over 10% are more likely to be on time.
The decision tree classifier was the most accurate (69%), outperforming the random forest classifier (67%), logistic regression (66%), and K Nearest Neighbors model (64%).