we define true-positives (tp) and true-negatives (tn) in
an analogous manner. With these definitions in place we
compute precision and re call and evaluate our algorithm
We choose the day October 2, 2020 to evaluate our
algorithm. We obtained for each customer that placed
a return request on the Myntra platform on this day the
repetition probability estimate and also product return
compliance status of the customer in the past 9 months.
We classified the customer accor ding to Algorithm 1
Our emphas is is on false -positives as it impacts re-
turned product resale value. Hence we aim to maximize
precision with a satisfactory recall. We note here that
precision is given by
tp
tp+fp
and recall is given by
tp
tp+fn
.
We compare d our algorithm aga inst classification
algorithms like logistic reg ression and random forests.
For these algorithms s ome of the important features are
summarised in Table 3. We note that our algo rithm
has better recall compared to logistic regression and
random forests and all algorithms report c lose to perfect
precision. We summarise the comparison in Table 2
We note that our algorithm is similar to a deci-
sion tree. Here, unlike splitting of a node in a classical
decision tr ee, we split the node based on eq uilibrium
strategy given by the game. Hence it may be possible
to fine tune hyper parameters of ensemble classific ation
algorithms, for e.g., of random forests to achieve better
performance. However unlike these classification algo-
rithms our algorithm is easily explainable - it is easy to
see the conditions under which a customer is elite, a de-
sirable feature for businesses in the context of customer
experience. In summary, the classific ation algorithm
based o n game theory methodologies achieves desired
precision with a satisfactory recall and has the addi-
tional advantage of explainability.
Our Logistic Random
Algorithm Regression Forests
Precisio n 99.7% 99.7% 99.6%
Recall 59.0% 48.4% 57.8%
Table 2: Comparis on of our classification algorithm
7 Conclusion
We modeled the pro ble m of identification of elite cus-
tomers and preferential returns processing for them as
a two-player repeated game and solved for its equilib-
rium strategy. We designed an algorithm based on this
analysis a nd evaluated the algorithm on real-world data
available at Myntra. We note that our algorithm is scal-
able, easy to implement and has performance compara-
ble to classification algorithms like logistic regression
and random forests. Moreover it has the advantage of
Features Description
#Q2 No. of complia nce failures
in last 9 months by customer
#Q1 No. of complia nce s uccesses
in last 9 months by customer
nserves No. of times the
returned product is served by Myntra
δ Estimated probability of
repetition of the game
l status Compliance state of
last return r equest by customer
Table 3: Important features o f the classifica tion algo-
rithms
explainability.
References
[1] Roberto Aringhieri, Davide Duma, and Vito Fragnelli.
Modeling the rational behavior of individuals on an e-
commerce system. Operations Research Perspectives,
5:22–31, 2018.
[2] Abd elsalam H Busalim, Fahad Ghabban, et al. Cus-
tomer engagement behaviour on social commerce plat-
forms: an empirical study. Technology in Society,
64:101437, 2021.
[3] Gabriele Camera, Marco Casari, and Maria Bigoni.
Cooperative strategies in anonymous economies: an ex-
periment. Games and Economic Behavior, 75(2):570–
586, 2012.
[4] David Gefen. Customer loyalty in e-commerce. Jour-
nal of the association for information systems, 3(1):2,
2002.
[5] Robert H Guttman an d Pattie Maes. Cooperative
vs. competitive multi-agent negotiations in retail elec-
tronic commerce. In International Workshop on Coop-
erative Information Agents, pages 135–147. Springer,
1998.
[6] Martin G Helander and Halimahtun M Khalid. Mod-
eling the customer in electronic commerce. Applied er-
gonomics, 31(6):609–619, 2000.
[7] Ralph L Keeney. The value of internet commerce to the
customer. Management science, 45(4):533–542, 1999.
[8] Hongzhen L ei, Di Lu, and Hon gHong Zhang. Research
on promotion of consumers’ ap plication of after ser-
vice in online shopping based on evolutionary game
theory–introduction of smart contract. In E3S Web of
Conferences, volume 233. EDP Sciences, 2021.
[9] Oskar Morgenstern and John Von Neumann. Theory
of games and economic behavior. Princeton university
press, 1953.
[10] Yadati Narahari. Game theory and mechanism design,
volume 4. World Scientific, 2014.
Copyright © 2021 by SIAM
Unauthorized reproduction of this article is prohibited