Overview / Usage

This project aims to study the sales pattern in a Late Night Restaurant, All Night Canteen (ANC) of BITS Pilani. The study provides insight into the student's purchase habits - thereby revealing products that are popular among the various student segments.

The governing body of ANC needs to increase their revenue to cater the rising wages of their employees. Also, it fears that the sales of less popular items (having low ratings) will fall in the coming months. The aim of this study is to hence develop policies that can improve the sales patterns.

We break this into two subproblems,

1. to implement a dynamic pricing scheme depending on the amount of sale for each hour of the day, and
2. to pair popular products with less popular ones to introduce combo offers for sale maximization.
   We also infer some interesting observations in the sales.

Methodology / Approach

1. For Dynamic Pricing Scheme, we use Association Rule Mining to create rules of the form:
   Hour, Student Segment -> Item ID.

This implies that, given an hour of the day, and the student type, he/she is likely to buy the item given by Item ID. Hence, if enough support and confidence is obtained for such an association, we increase the price of this item to an appropriate level for that hour.

We consider all such triplets, and analyze the support and confidence measures of this kind of association. We apply frequent item set mining and vary the minimum support (minsup) and minimum confidence values to obtain an appropriate increase in profits.

Here, an appropriate penalty measure is taken to penalize rising profits. Our goal is to maximize profits while keeping the penalty minimum. As a threshold, we consider 5% to be an appropriate rise in profits.

1. For pairing popular products with less popular ones, we consider associations of the form:
   Popular product -> non Popular product

This implies that when a customer buys a popular product, he/she is likely to buy the non popular item. If we can bundle these two products and give a combo offer (offering a discount) we can rise the sales of both the items. Hence, increased prices from the dynamic pricing scheme in 1 may be compensated by such offers, thereby increasing sales of those whose sales may fall drastically.

Further details and code is uploaded in the github repository.

Technologies Used

Python is used for data pre-processing and Association Rule Mining. We develop the code from scratch.

Repository

https://github.com/nmakes/ANC-data-mining