Abstract
Context-aware ubiquitous computing systems should be able to introspect the surrounding environment and adapt their behavior according to other existing systems and context changes. Although numerous ubiquitous computing systems have been developed that are aware of different types of context such as location, social situation, and available computational resources, few are aware of their computational behavior. Computational behavior introspection is common in reflective systems and can be used to improve the awareness and autonomy of ubicomp systems. In this paper, we propose a decentralized approach based on Simple Network Management Protocol (SNMP) and Universal Plug and Play (UPnP), and on state transition models to model and expose computational behavior. Typically, SNMP and UPnP are targeted to retrieve raw operational variables from managed network devices and consumer electronic devices, e.g., checking network interface bandwidth and automating device discovery and plug and play operations. We extend the use of these protocols by exposing the state of different ubicomp systems and associated state transitions statistics. This computational behavior may be collected locally or remotely from ubicomp systems that share a physical environment, and sent to a coordinator node or simply shared among ubicomp systems. We describe the implementation of this behavior awareness approach in a home health-care environment equipped with a VoIP Phone and a drug dispenser. We provide the means for exposing and using the behavior context in managing a simple home health-care setting. Our approach relies on a system state specification being provided by manufacturers. In the case where the specification is not provided, we show how it can be automatically discovered. We propose two machine learning approaches for automatic behavior discovery and evaluate them by determining the expected state graphs of our two systems (a VoIP Phone and a drug dispenser). These two approaches are also evaluated regarding the effectiveness of generated behavior graphs.

Abstract

Abstract
In this work, a cross-validation procedure is used to identify an appropriate Autoregressive Integrated Moving Average model and an appropriate state space model for a time series. A minimum size for the training set is specified. The procedure is based on one-step forecasts and uses different training sets, each containing one more observation than the previous one. All possible state space models and all ARIMA models where the orders are allowed to range reasonably are fitted considering raw data and log-transformed data with regular differencing (up to second order differences) and, if the time series is seasonal, seasonal differencing (up to first order differences). The value of root mean squared error for each model is calculated averaging the one-step forecasts obtained. The model which has the lowest root mean squared error value and passes the Ljung-Box test using all of the available data with a reasonable significance level is selected among all the ARIMA and state space models considered. The procedure is exemplified in this paper with a case study of retail sales of different categories of women's footwear from a Portuguese retailer, and its accuracy is compared with three reliable forecasting approaches. The results show that our procedure consistently forecasts more accurately than the other approaches and the improvements in the accuracy are significant.

Abstract
Forecasting future sales is one of the most important issues that is beyond all strategic and planning decisions in effective operations of retail supply chains. For profitable retail businesses, accurate sales forecasting is crucial in organizing and planning purchasing, production, transportation and labor force. Retail sales series belong to a special type of time series that typically contain strong trend and seasonal patterns, presenting challenges in developing effective forecasting models. This paper compares the forecasting performance of state space models and ARIMA models. The forecasting performance is demonstrated through a case study of retail sales of five different categories of women footwear: Boots, Booties, Flats, Sandals and Shoes. An approach based on cross-validation is used to identify automatically appropriate state space and ARIMA models. The forecasting performance of these models is also compared by examining the out-of-sample forecasts. The results indicate that the overall out-of-sample forecasting performance of ARIMA models evaluated via RMSE, MAE and MAPE is better than state space models. The performance of both forecasting methodologies in producing forecast intervals was also evaluated and the results indicate that ARIMA produces slightly better coverage probabilities than state space models for the nominal 95% forecast intervals. For the nominal 80% forecast intervals the performance of state space models is slightly better.

Abstract
Aggressive marketing causes rapid changes in consumer behavior and some significant impact in the retail business. In this context, the sales forecasting at the SKU level can help retailers to become more competitive by reducing inventory investment and distribution costs. Sales forecasts are often obtained combining basic univariate forecasting models with empirical judgment. However, more effective forecasting methods can be obtained by incorporating promotional information, including price, percentage of discount (direct discount or loyalty card discount), calendar events and weekend indicators not only from the focal product but also from its competitors. To deal with the high dimensionality of the variable space, we propose a two-stage LASSO regression to select optimal predictors and estimate the model parameters. At the first stage, only focal SKUs promotional explanatory variables are included in the Autoregressive Distributed Lag model. At the second stage, the in-sample forecast errors from the first stage are regressed on the explanatory variables from the other SKUs in the same category with the focal SKU, and to use that information more effectively three different approaches were considered: select the five top sales SKUs, include all raw promotional information, and preprocess raw information using Principal Component Analysis. The empirical results obtained using daily data from a Portuguese retailer show that the inclusion of promotional information from SKUs in the same category may improve the forecast accuracy and that better overall forecasting results may be obtained if the best model for each SKU is selected.

Abstract
Sales forecasts gained more importance in the retail industry with the increasing of promotional activity, not only because of the considerable portion of products under promotion but also due to the existence of promotional activities, which boost product sales and make forecasts more difficult to obtain. This study is performed with real data from a Portuguese consumer goods retail company, from January 2012 until April 2015. To achieve the purpose of the study, dynamic regression is used based on information of the focal product and its competitors, with seasonality modelled using Fourier terms. The selection of variables to be included in the model is done based on the lowest value of AIC in the train period. The forecasts are obtained for a test period of 30 weeks. The forecasting models overall performance is analyzed for the full period and for the periods with and without promotions. The results show that our proposed dynamic regression models with price and promotional information of the focal product generate substantially more accurate forecasts than pure time series models for all periods studied.