Explore your business data.

Association Algorithms

In terms of transactional data, we are interested in analyzing:

◆ How are shopping baskets made up?

◆ What is the adoption sequence for products? Example of product 1: credit card, and product 2: savings account.

◆ What are the sequences of steps a user follows on a webpage? Find the rules capable of describing the association of consumer decisions, shopping baskets, browsing routines, patterns of cross-selling and staggered selling; these are just some of the applications of association algorithms.

Cluster Analysis

Our goal is to find groups (customers, points of sale, websites).

The elements classified within these groups are very much alike but, when comparing between groups, these must behave very differently.

Clustering techniques are generally used to find segments, natural groupings, profiles, etc.

Segmentation of customers, risks, contracts, etc., are some of the applications.

Reducing Dimensions

When faced with many variables, it is common to find that these tend to be correlated.

Our goal is to build new variables or vectors that group correlated variables and that make future forecasts more accurate and stable.

Techniques such as principal component analysis, factor analysis and multidimensional scaling are some of the examples.

Self-organized Maps

This family of tools comes from artificial neural networks.

It seeks to find segments and ultimately profiles, revealing hidden patterns in the information.

Based on discovering close neighbors in space or in time, which is why this is an efficient technique when the analysis units tend to vary through time or topologically

Regression methods

These models generate equations where the X variables (explanatory) acquire relevance and can then estimate the Y variable (factor we want to predict).

Some models in this family of techniques are:

◆ Linear Regression: To predict quantitative variables.

◆ Logistic Regression: To predict binary or multinomial qualitative variables.

◆ Poisson Regression: To predict variables that are expressed as percentages or tallies.

◆ Cox Regression: To predict durations.

Tree methods

Decision trees build partitions in the data according to the impact of X variables (explanatory) on the Y variable.

It creates classification rules, not equations.

Some algorithms are CHAID, CRT, Entropies.

Some advanced machine learning techniques are based on this family of techniques. Example: Random Forest, Gradient boosting.

Neural Networks methods

The neural network tries to learn on Y based on the X variables available, the same way the brain tries to establish neural connections to generate recognition and memory.

Equations and rules are created simultaneously to achieve the prediction.

The multilayer perceptron and the radial basis function are two of the most frequently used algorithms to train models of this nature.

Smoothing Methods

These are efficient methodologies to model the historic behavior of a variable, but not so optimal to generate its future representation.

These methods tend to be efficient when there is little data.

Moving averages, exponential smoothing, Holt line, Winters method.

Box-Jenkins Method

Also known as ARIMA, is a comprehensive methodology to decompose the series based on three elements:

a. AR: Autoregressive component.

b. I: Integrated component – differentiation or seasonality.

c. MA: Moving average component.

If the three components predict the series, an ARIMA is established; when only components AR and MA are significant, an ARMA is established, etc.

Neural Networks Methods

Neural networks have shown excellent accuracy levels to predict time series.

Especially the radial basis function models, which position themselves as a new strategy to improve the forecast accuracy of forecast models.

Transfer and Co-integration functions

We may come across opportunities where we have different variables deployed over time and on the same dates.

In these cases, we want to construct integrative models called transfer functions.

Our goal is to analyze how a time series predicts or estimates another one.

Mathematical Optimization

The relationship between processes and results; they tend to assume a behavior that describes these relationships.

Linear, non-linear, quadratic optimization, etc.

Discrete Event Simulation

Qualitative results are also results and we need to predict them, estimate them and take them into account in order to optimize them.

Optimization of Resources and Processes

Throughout productive processes, when managing projects or in similar scenarios, time and money need to be strategically allocated.

Resource assignment to achieve the best result with less represents the most frequent optimization issues in companies.

Design and Optimization of Routes

Moving things – persons from one point to another, as quickly as possible, with the highest feasible occupation and at the lowest cost, is a challenge.

Operations research has an efficient answer to this type of questions.

Video processing

Isolation of graphic patterns on videos to alert on outcomes of interest.

Multidimensional analysis: Shape, color, depth, speed, etc., to discover findings and generate applications.

Sound processing

Recognition of voice, words and establishment of patterns.

Training bots and use of this knowledge / insight to develop business solutions and applications.

Photo processing

Isolating graphic patterns of images, face recognition and development of predictive applications within said dimension.