Modules and Technologies

SAMV - classifier-adversarial boosting

Designed to increase the efficiency of data separation of objects set by the competitive error prediction method.

Module for cleaning outliers by changing feature space

A quick way of cleaning outliers in data across different distribution types. Does not require manual assignment of clearing trashholds. From our experience, 99% of the data requires fields to be cleared of outliers.

Bulk purchaser detection module

Allows you to identify and mark wholesale buyers automatically by econometric indicators, typical sales structure and business type. It is necessary to eliminate the influence of wholesale buyers and wholesale outlets on market performance, as well as to eliminate duplication of sales.

Module of completing data by bounded sampling

Allows you to drastically improve the quality of data in sampling that are limited in size. The ability to increase granularity is one of the consequences of this module.

Calibration and data improvement module based on the gravitational model of the calibration kernel.

Calibration to the general population (GP) and data representativeness is one of the main analysis aspirations and request of many clients. The module consists of a group of submodules for assessing the quality of sample objects, representing and constructing a frame of GP, placing objects and calibrating weights based on a gravity model of object quality. Allows you to automatically calculate the objects weights to calibrate the sample to GP.

Rotation compensation module

We often work with highly rotated samples. The calculation of indicators on such samples always requires a special approach. This module is designed to stabilize the sample by monitoring an object life cycle in the sample, assessing its quality and characteristics, as well as replacing a retired object with the most similar one.

Data preparation module for ML "Double normalization"

Allows you to reduce the multidimensional consumption of sample objects to a vector of coefficients even in the abnormality case and skewed consumption

Module of determining connections in sales dynamics

Identifies the reasons for the sales change through the change of basis. Allows you to evaluate the influence of factors on sales for a particular category/SKU.

Anomaly Detection Technique with ML

Automation of anomaly detection with minimization of human involvement.

Adaptive RIM Weighting Method

Our improvement to the popular iterative weighting method, increasing the accuracy of the weighting factors.

The method of extrapolation filling the gaps "Avatar"

Allows you to reduce or eliminate the impact of losses in specific types of data. We use it in panel data with different levels of confidence to respondent, different quality of respondents, and in channels with an unstable signal from the the sampling objects under study.

Iterative Clustering User Profiling Technique

Automatic performance of creative tasks of identifying features and clusters with little or no human involvement.

Methodology for determining competitors by similarity

Allows you to identify real competitors to a brand or product. Not those that simply appear in the buying structure or are positioned as competitors, but those that have the greatest chance of replacing the target product when its position in the market decreases.