Abstract: Techniques described herein can monitor various data metrics. The techniques can select a subset of dimensions from a plurality of dimensions related to a data shift. The techniques including generating a plurality of decision tree graphs to classify a plurality of segments, each segment representing a combination of two or more dimensions of the subset of dimensions, and each decision tree graph including a different root node representing a respective dimension of the subset of dimensions.

Abstract: Disclosed is a fast and accurate time series forecasting algorithm that eliminates the need for hyperparameter tuning. Time series data may be analyzed using a quadratic function to determine a quadratic trend prediction, which is removed from the time series data to generate first detrended time series data. A moving median of the time series data is determined and the moving median is removed from the time series data to generate second detrended time series data. An amplitude scaling factor is determined based on the second detrended time series data and the first detrended time series data is descaled using the amplitude scaling factor to generate descaled time series data. The descaled time series data is analyzed to determine a seasonal prediction and a time series forecast is generated based on the seasonal prediction, the quadratic trend prediction, and the amplitude scaling factor.

Abstract: Using an attributes model of a time series forecasting model, determine a set of features based on time series data, the set of features including periodic components. The time series data may be divided into a set of segments. For each segment of the set of segments, a weight may be assigned using an age of the segment, resulting in a set of weighted segments of time series data. Using a trend detection model of the time series forecasting model, trend data from the set of weighted segments of time series data may be determined. A time series forecast may be generated by combining the set of features and the trend data.

Abstract: Using an attributes model of a time series forecasting model, determine a set of features based on time series data, the set of features including periodic components. The time series data may be divided into a set of segments. For each segment of the set of segments, a weight may be assigned using an age of the segment, resulting in a set of weighted segments of time series data. Using a trend detection model of the time series forecasting model, trend data from the set of weighted segments of time series data may be determined. A time series forecast may be generated by combining the set of features and the trend data.

Abstract: A processing device may analyze a set of time series data using a time series forecasting model comprising an attributes model and a trend detection model. The attributes model may comprise a modified gradient boosting decision tree (GBDT) based algorithm. Analyzing the set of time series data comprises determining a set of features of the set of time series data, the set of features including periodic components as well as arbitrary components. A trend of the set of time series data may be determined using the trend detection model and the set of features and the trend may be combined to generate a time series forecast.

Abstract: Disclosed are methods and apparatus for updating at least one prediction model for use by at least one interactive server. Each interactive server performs a plurality of actions in the context of a plurality of input attributes values of an input dataset wherein the actions are selected based on each prediction model. contextual data are automatically and continually obtained from the interactive server as it performs the plurality of actions. A learning model is automatically and continually updated based on all of the obtained contextual data. An updated prediction model and a prediction of a probability of an outcome using the updated prediction model are generated. A prediction is generated and an action of the plurality of actions based on the prediction is selected. The selection action is performed.

Abstract: An expected value associated with an end event is predicted for each of multiple event sequences, with an event sequence to be presented to a user being selected based on the excepted values. In one embodiment, input attributes and contextual data collected during performance of previous event sequences are obtained, and a plurality of expected values is predicted for going from a first event to each of a plurality of subsequent events of a first event sequence. An expected value is also predicted for going from a first event of a second event sequence to an end event of the second event sequence. The expected values for the first and second event sequences are compared to determine which end event a user is more likely to reach, and an event sequence is selected to present to a user based thereon.

Abstract: Disclosed are methods and apparatus for reporting significant data mining changes. In general, embodiments of the present invention address the shortcomings of the prior art through comparison over time of prediction model characteristics, such as inferences. Embodiments of the present invention detect trends in the model itself by detecting changes in levels of correlation (or any other model aspect) between individual elements of input data and targets of predictions. In this specific embodiment, users of the model are preferably alerted when an input characteristic or other model aspect, which was not important before, becomes important and when an input characteristic, which was important, loses its importance.

Abstract: Disclosed are methods and apparatus for optimizing results produced by a predictive model in order to determine which action to perform out of a plurality of actions. In an operation (a), a plurality of goal metrics are provided for a plurality of possible actions based on a plurality of input conditions. One or more of the goal metrics are produced by one or more predictive models. In an operation (b), the plurality of goal metrics are normalized. In an operation (c), for each possible action a total of each of the normalized goal metrics multiplied by a corresponding predetermined weight is determined. In an operation (d), the totals determined for the plurality of possible actions are compared to thereby determine a highest total. In an operation (e), an action selected from the plurality of possible actions is performed, where the selected action has the highest total.

Abstract: Disclosed are methods and apparatus for updating (i.e., generating or modifying) one or more prediction models that are used to make decisions as to which interactions (such as which automated voice option to present to a person who has contacted an automated telephone service center) to perform under a specified range of input conditions (such as the calling person's profile, which interactions have taken place so far between the call center and the calling person, etc.). In general terms, the present invention provides a feedback mechanism for updating at least one prediction model based on contextual data that is continuously collected during interaction processes (such as various telephone interactions with a telephone service center). The updating of the prediction model includes pruning inputs that are statistically insignificant from the prediction model.

Abstract: Disclosed are methods and apparatus for evaluating a certainty characteristic of a predictive model. When a decision needs to be implemented, the predictive model is utilized unless the certainty characteristic of such model indicates that the predictive model results are unacceptably uncertain and should not be used. Otherwise, the predictive model is used to reach a decision. In a further embodiment, randomization is introduced into the results of the predictive model (when utilized for a decision). The amount of randomization is tied to the amount of uncertainty of results of the model to thereby balance exploitation and exploration goals.

Abstract: Disclosed are methods and apparatus for predicting an expected value associated with an end event of an event sequence. In one embodiment, the following operations are performed: (a) providing a current set of input attributes and contextual data collected during performance of previous event sequences; (b) predicting a plurality of expected values for going from a first event of a known event sequence to each of a plurality of subsequent events of the known event sequence based at least on the current set of input attributes and the collected contextual data; and (c) predicting an expected value for going from a first event of an unknown event sequence to an end event of such unknown event sequence based on at least two of the expected values predicted for the known event sequence and based at least on the current set of input attributes and the collected contextual data.