Multivariate text mining for process improvement using cross-canonical correlation analysis

Text analysis is a useful tool to determine what a company and its customers want in order to improve processes and methodologies of analysis. Searches in databases may have a time series component that determines the importance and sequences of multivariate searches and its structure. This paper presents a methodology to simplify and model multivariate searches in time using the Canonical Correlation approach. The techniques shown provide a robust methodology to simplify the analysis and create predictive models taking into account temporal dependencies.

An Analysis of Lost Sales

The purpose of this manuscript is to shed light on problems associated with lost sales and the incurring of cost associated with lost sales. An investigation is made to determine if seasonality in sales and lost sales have effects on the efficient operations of supply chains. Optimization is always a goal of management supply chains, but cost increases due to insufficient inventory, low-quality product and the like lead to customers not returning. These are lost sales that occur for many reasons. We study a data set to determine if the ignoring of time series component also has an effect on the variation in lost sales. If so, can we measure the magnitude of the effects of seasonal variation in lost sales, and what are their consequences?

Data analysis toolkit for long-term, large-scale experiments

A new data analysis toolkit which is suitable for the analysis of large-scale, long-term datasets and the phenomenon/anomalies they represent is described. The toolkit aims to expose and quantify scientific information in a number of forms contained within a time-series based dataset in a quantitative and rigorous manner, reducing the subjectivity of observations made, thereby supporting the scientific observer. The features contained within the toolkit include the ability to handle non-uniform datasets, time-series component determination, frequency component determination, feature/event detection and characterization/parameterization of local behaviours. An application is presented of a case study dataset arising from the 'Lasgit' experiment.

Time‐Series Cross‐Section Methods

This article outlines the literature on time-series cross-sectional (TSCS) methods. First, it addresses time-series properties including issues of nonstationarity. It moves to cross-sectional issues including heteroskedasticity and spatial autocorrelation. The ways that TSCS methods deal with heterogeneous units through fixed effects and random coefficient models are shown. In addition, a discussion of binary variables and their relationship to event history models is provided. The best way to think about modeling single time series is to think about modeling the time-series component of TSCS data. On the cross-sectional side, the best approach is one based on thinking about cross-sectional issues like a spatial econometrician. In general, the critical insight is that TSCS and binary TSCS data present a series of interesting issues that must be carefully considered, and not a standard set of nuisances that can be dealt with by a command in some statistical package.