Business Objects Software Limited

Paul O'Hara

Malte Christian Kaufmann

Ying Wu

Esther Rodrigo Ortiz

Alan McShane

Anirban Banerjee

Orla Cullen

Artur Stulka

Conor White

Paul O'Connor

Paul Pallath

Paul Sheedy

Cathal McGovern

Michael Golden

Vlad Zat

Satinder Singh

Mairtin Keane

Gabor Szabo

Ian Denver

Jiazheng Li

Kerry O'Connor

Mark Ahern

Ben Murphy

Mateusz Lewandowski

Emmet Norton

Robert McGrath

Anomalies may be detected using a multiple machine learning model anomaly detection framework. A clustering model is trained using an unsupervised machine learning algorithm on a historical anomaly dataset. A plurality of clusters of records are determined by applying the historical anomaly dataset to the clustering model. Then it is determined whether each cluster of the plurality of clusters is an anomaly-type cluster or a normal-type cluster. The plurality of labels for the plurality of records are updated based on the particular record's cluster classification. Non-pure clusters are determined from among the plurality of clusters based on a purity threshold. A supervised machine learning model is trained for each of the non-pure clusters using the records in the given cluster and the labels for each of those records. Then, predictions of an anomaly are made using the clustering model and the supervised machine learning models.

Multiple machine learning model anomaly detection framework

Using approximated bin intervals to label the histograms provides clarity and allows for the histogram to be more intuitively understood. A dataset may comprise a plurality of records having a plurality of features including one or more continuous features. A selection of a continuous feature may be obtained. A bin width based on a number of bins and feature statistics of the continuous feature may be determined. An approximated bin interval range is determined by applying a bin mask based on the bin width to the feature statistics. An approximated bin width is determined based on the number of bins and the approximated bin interval range. Approximated bin intervals for the histogram are determined based on the approximated bin width. A histogram is generated having bins with intervals based the approximated bin intervals.

Histogram bin interval approximation

The present disclosure provides for accurate and efficient identification of candidate features for an input dataset comprising one or more continuous features and one or more categorical features is obtained. A number of categorical feature categories based on the one or more categorical features is determined. Record counts for each of the categorical feature categories are determined. Skew statistics for each category are determined based on the record counts for each of the categorical feature categories. Cardinality skew factors for each of the one or more categorical features are then determined based on the record counts and the skew statistics. A number of the one or more categorical features having the highest cardinality skew factors are selected from among the cardinality skew factors. Then, a top contributor deviation analysis is performed using the selected number of the categorical features having the highest cardinality skew factors.

Feature selection for deviation analysis

A system and method include input of data records to a first trained predictive model to obtain a predicted value associated with each input data record. A model region is then associated with each of the input data records based on the first trained predictive model, the input data records and the predicted values. Enhanced input data records are generated by, for each model region, adding derived values of engineered features associated with the model region to input data records associated with the model region and default values of the engineered features associated with the model region to input training records not associated with the model region. The enhanced input data records are input to a second trained predictive model to obtain an enhanced predicted value associated with each input data record.

Framework for providing improved predictive model

The present disclosure involves systems, software, and computer implemented methods for automatically detecting hot areas in heat map visualizations. One example method includes identifying a two-dimensional heat map. The identified two-dimensional heat map is converted to a one-dimensional heat map. Cells of the one-dimensional heat map are clustered using a density-based clustering algorithm to generate at least one dense region of cells. A mean value of cells in each dense region is calculated and the dense regions are sorted by mean value in descending order. An approach for identifying hot areas is selected and the selected approach is used to identify at least one dense region as a hot area of the one-dimensional heat map.

Automatic hot area detection in heat map visualizations

Systems and methods include determination, for each of a plurality of discrete features, of statistics based on a number of occurrences of each discrete value of the discrete feature in the data, determination of first summary statistics based on the determined statistics, determine of a dissimilarity for each discrete feature based on the first summary statistics and on the statistics determined for the discrete feature, determination of candidate discrete features based on the determined dissimilarities, determination, for each of the candidate discrete features, of second summary statistics based on values of a continuous feature associated with each discrete value of the candidate discrete feature, determination of a deviation score for each of the candidate discrete features based on the second summary statistics, and transmission of the candidate discrete features for display in association with the continuous feature based on the determined deviation scores.

Continuous feature-independent determination of features for deviation analysis

Systems and methods include reception of a set of data including continuous features and a discrete feature, each continuous feature associated with a plurality of values and the discrete feature associated with a plurality of discrete values, determine, for each continuous feature, a relationship factor representing a relationship between the discrete feature and the continuous feature based on the plurality of values associated with the continuous feature and the plurality of discrete values, identify one of the continuous features associated with a largest one of the determined relationship factors, generate, for each of the other features, a correlation factor representing a correlation between the continuous feature and the identified continuous feature, determine, for each of the continuous features other than the identified continuous feature, a composite relationship score based on the relationship factor and the correlation factor associated with the feature, and present a visualization associated with the discrete feature, the identified continuous feature, and a continuous feature associated with a largest composite relationship score.

Composite relationship discovery framework

Systems and methods include determination, determine, for each of a plurality of discrete features, of statistics for each discrete value of the discrete feature based on values of a continuous feature associated with the discrete value, determination, for each discrete feature, of first summary statistics based on the statistics determined for each discrete value of the discrete feature, determination, for each discrete feature, of a dissimilarity based on the first summary statistics determined for the discrete feature and on the statistics determined for each discrete value of the discrete feature, determination of candidate discrete features of the discrete features based on the determined dissimilarities, the candidate discrete features comprising less than all of the discrete features, determination, for each of the candidate discrete features, of second summary statistics based on values of the continuous feature associated with each discrete value of the candidate discrete feature, determine of a deviation score for each of the candidate discrete features based on the second summary statistics, and presentation of the candidate discrete features based on the determined deviation scores.

Determination of candidate features for deviation analysis

The present disclosure involves systems, software, and computer implemented methods for ranking time dimensions. One example method includes receiving a request for an insight analysis for a dataset that includes a value dimension and a set of multiple date dimensions. Each date dimension is converted into a time series and a value quality factor is determined for each time series that represents a level of data quality for the time series. A time series informative factor is determined for each time series that represents how informative the time series is within a specified time window. An insight score is determined, for each time dimension, based on the determined value quality factors and the determined time series informative factors. The insight score for the time dimension is provided, for at least some of the time dimensions.

Flexible algorithm for time dimension ranking

The present disclosure involves systems, software, and computer implemented methods for automatically recommending one or more frequencies for time series data. One example method includes receiving a request for an insight analysis for an input time series included in a dataset. For each of multiple frequencies to analyze, the input time series is transformed into a frequency time series. An absolute percentage change impact factor and an absolute trend impact factor are determined for each frequency time series. A frequency interest score is determined based on the determined absolute percentage change factors and the determined absolute trend impact factors, for each time frequency time series. The frequency interest score is provided for at least some of the frequency time series.

Automatic frequency recommendation for time series data

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, to predict future Day Sales Outstanding (DSO) forecasts for a number of future time periods. In one aspect, a method includes receiving open receivables financial line item data and revenue financial line item data, providing the open receivables financial line item data to a DSO predictor engine to generate a predicted open receivables that includes a multi-step time series forecasting regression generated from the open receivables financial line item data, providing the revenue financial line item data to the DSO predictor engine to generate a predicted revenue comprising the multi-step time series forecasting regression generated from the revenue financial line item data; generating a predicted DSO with the predicted open receivables and predicted revenue, and providing the predicted DSO to a client device.

Multi-step day sales outstanding forecasting

A method includes receiving a plurality of items, grouping the plurality of items into a plurality of clusters, where each of the plurality of clusters comprises items having similar features to one another, applying a classification model to each cluster to predict whether each item of a cluster will be delivered on time or delivered late, applying a regression model that determines an expected measure of tardiness of each item predicted to be delivered late, and outputting a delivery date prediction for each item predicted to be delivered late based on the expected measure of tardiness of the item.

Growing community of inventors

Dublin, Ireland

Paul O'Hara