International Business Machines Corporation

Sier Han

Jing-Yun Shyr

Yea Jane Chu

Jing James Xu

Damir Spisic

Chao Zhang

Xiao Ming Ma

Ming Xie

A Peng Zhang

Chun Hua Tian

Ning Sun

Feng Juan Wang

Xueying Zhang

Yea-Jane Chu

Xiu Fang Zhu

Lei Fan

Dong Chen

Jing Jing Zhang

Graham John Wills

Ji Hui Yang

Lei Gao

Fan Li

Weicai Zhong

Long Jiao

Zongyao Zhang

Embodiments include predicting transactions by an entity and identifying promotions to offer the entity. Aspects include parsing a plurality of event records corresponding to a plurality of entities respectively. Aspects also include identifying a sequence of events corresponding to the entity and discretizing time durations and event values of the sequence of events into discrete symbolic values. Aspects further include generating a temporal pattern of events in the sequence of events, the temporal pattern including a sequence of transaction-symbols representative of the time duration and the event value of the events in the sequence of events of the entity and predicting a next transaction based on the temporal pattern.

Transaction data analysis

Embodiments include predicting transactions by an entity and identifying promotions to offer the entity. Aspects include parsing a plurality of event records corresponding to a plurality of entities respectively. Aspects also include identifying a sequence of events corresponding to the entity and discretizing time intervals and event values of the sequence of events into discrete symbolic values. Aspects further include generating a temporal pattern of events in the sequence of events, the temporal pattern including a sequence of transaction-symbols representative of the time interval and the event value of the events in the sequence of events of the entity and predicting a next transaction based on the temporal pattern.

Statistically significant event patterns predict the timing for performing entity maintenance. Event patterns are determined based on a target variable having an undesired value for a given entity when the event pattern occurs. Event patterns are filtered based on distributions of the event patterns across multiple entities and distributions of event patterns during desired operation of the entities and undesired operation of the entities. A predictive maintenance process is established having significant event patterns as the basis for maintenance tasks.

Predictive maintenance utilizing supervised sequence rule mining

Updating a prediction model, where the prediction model is used for time series data, a computer selects a first prediction time window in an order from a plurality of prediction time windows associated with the prediction model, and predicts predicted values of the time series data at time points within the first prediction time window. The computer calculates a prediction error associated with the first prediction time window based on the one or more predicted values and one or more actual measured values of the time series data at the plurality of time points. The computer determines whether the prediction error is larger than a predefined error threshold associated with the first prediction time window, and in response to determining the prediction error is larger than the predefined error threshold, provides a notification of updating the prediction model.

Updating prediction model

Techniques are described for generating characterizations of time series data. In one example, a method includes extracting a trend-cycle component, a seasonal component, and an irregular component from a time series of data. The method further includes performing one or more pattern analyzes on the trend-cycle component, the seasonal component, and the irregular component. The method further includes, for each pattern analysis of the one or more pattern analyzes, performing a comparison of an analytic result of the respective pattern analysis to a selected significance threshold for the respective pattern analysis to determine if the analytic result passes the significance threshold for the respective pattern analysis. The method further includes generating an output for each of the analytic results that pass the significance threshold for the respective pattern analysis.

Automatic time series exploration for business intelligence analytics

Techniques are described for generating characterizations of time series data. In one example, a method includes extracting a trend-cycle component, a seasonal component, and an irregular component from a time series of data. The method further includes performing one or more pattern analyses on the trend-cycle component, the seasonal component, and the irregular component. The method further includes, for each pattern analysis of the one or more pattern analyses, performing a comparison of an analytic result of the respective pattern analysis to a selected significance threshold for the respective pattern analysis to determine if the analytic result passes the significance threshold for the respective pattern analysis. The method further includes generating an output for each of the analytic results that pass the significance threshold for the respective pattern analysis.

An approach for discovery of relevant data in massive datasets. Compare datasets including compare key fields, compare data fields and a core dataset including target data field(s) and core field(s) are received. The compare datasets are categorized into direct and indirect related dataset pools based on the target data field(s) correlation strength with matching compare and core fields. The direct related dataset pool and the core dataset are transformed into reduction datasets based on statistical measure of values of target data fields, shared key fields and compare data fields. Target correlations of the reduction datasets are creating based on a reduction compare and target data fields. Statistical relationship strength of core dataset and the direct related dataset pool are created based on a statistical mean of target correlations and a relevancy data store is created.

Automatic discovery of relevant data in massive datasets

Provided are techniques for imputing a missing value for each of one or more predictor variables. Data is received from one or more data sources. For each of the one or more predictor variables, an imputation model is built based on information of a target variable; a type of imputation model to construct is determined based on the one or more data sources, a measurement level of the predictor variable, and a measurement level of the target variable; and the determined type of imputation model is constructed using basic statistics of the predictor variable and the target variable. The missing value is imputed for each of the one or more predictor variables using the data from the one or more data sources and one or more built imputation models to generate a completed data set.

Missing value imputation for predictive models

Provided are techniques for interaction detection for generalized linear models. Basic statistics are calculated for a pair of categorical predictor variables and a target variable from a dataset during a single pass over the dataset. It is determined whether there is a significant interaction effect for the pair of categorical predictor variables on the target variable by: calculating a log-likelihood value for a full generalized linear model without estimating model parameters; calculating the model parameters for a reduced generalized linear model with a recursive marginal mean accumulation technique using the basic statistics; calculating a log-likelihood value for the reduced generalized linear model; calculating a likelihood ratio test statistic using the log-likelihood value for the full generalized linear model and the log-likelihood value for the reduced generalized linear model; calculating a p-value of the likelihood ratio test statistic; and comparing the p-value to a significance level.

Interaction detection for generalized linear models for a purchase decision

A subset of (k&#x2212;1)-dimensional tables are received, wherein k is greater than 1. A set of k-dimensional tables is created by combining each of the (k&#x2212;1)-dimensional tables with a non-included dimension corresponding to a 1-dimensional table. Significance of interaction and interaction effect size is computed for the created set of k-dimensional tables to determine dimension and measure interactions.

Relationship discovery in business analytics

Provided are techniques for generating order statistics and error bounds. For each of multiple, distributed data sources, a finite number of data bins are created for each field in that data source. Data values in each of the multiple, distributed data sources are processed to generate basic summaries for each of the data bins in a single pass of the data values. The data bins from each of the multiple, distributed data sources are sorted. One or more approximate order statistics are computed for a data set by accumulating counts from a number of the sorted data bins. Lower and upper error bounds are provided for each of the computed one or more approximate order statistics, wherein the lower and upper error bounds are values delimiting an interval containing a true value of an order statistic.

Growing community of inventors

Xi'an, China

Sier Han