Time-Series Techniques for Prediction Sales in Cashback Websites
Keywords:
Cashback, business intelligence, data mining, forecastinghistorical data, websitesAbstract
Forecasting sales using time-series data has been an area of active research. In recent years, a significant amount of research effort went into forecasting sales in e-commerce platforms. The mechanisms of cashback are the main attractions for both advertisers and publishers. This B2C transaction enables the cashback website’s publishers to earn a commission, and from that, they offer a percentage back to the consumers/end users in the form of cashback (aka rewards). This paper proposes a model that leverages an existing time-series technique seasonal persistence for forecasting sales in cashback websites. This paper contribution is to select different time-series models like Arima, XGBoost, Fbprophet, and applied them on three popular European-based cashback websites Cashback Korting, Nucash, iPay, to forecast their sales. The results confirm that XGBoost performs comparatively better than the other selected models as it produces reasonably low error rates (e.g., the Mean absolute percentage error (MAPE) rate in XGBoost remains between 7 to 14). The evaluation results demonstrate that our proposed model produces reasonably low error rates (e.g., the MAPE rate of our model is under 1) and also requires less expertise and dimensions of data to forecast sales, which may further help in making effective business decisions and offer exciting propositions to drive the business forward.