D4TimeX

Technology Overview

D4TimeX is a time series analysis service, based on Machine Learning models and algorithms, designed by Dhiria, with advanced features for descriptive, predictive, and multivariate analysis, available in an as-a-service model through the D4 platform developed by Dhiria.

The key features of time series analyses produced by D4TimeX are as follows:

• Descriptive Analysis: Its goal is to characterize the behavior of the selected products by extracting knowledge and insights from the associated time series. In this functionality, D4TimeX can study periodicities, seasonality, and trends using tools such as Box-Plots and decomposition methods. Additionally, D4TimeX can measure autocorrelation and stationarity properties, which are key aspects in explaining time series behavior. The purpose of this descriptive analysis mode is to automatically extract all the information needed to understand, evaluate, and validate the sales behavior of specific products by analyzing historical series; 

• Predictive Analysis: Its goal is to learn and subsequently apply specific predictive models for time series forecasting. The key aspect of this functionality is the ability to automatically identify the subset of past data to predict what will happen in the future, thereby improving prediction accuracy. This feature uses various families of models developed by Dhiria, such as supervised models based on regression, statistical models, and neural network-based predictive models. The result of this phase is the prediction, provided by the best predictive models specific to each product in the dataset, along with a numerical evaluation of the prediction's quality (measured as prediction accuracy and mean squared error) for a time window ranging from one step ahead to six steps ahead;

• Multivariate Analysis: Its goal is to explore the correlations present in time series through the analysis of causal relationships in the data. These relationships are functional tools for both enhancing descriptive analysis and improving predictive analysis. The result of this operational mode is the creation of a dependency graph that can characterize and model the relationships between different time series.

D4TimeX provides the results of all these analyses both in numerical data form and as interactive charts that allow for visual exploration, with the ability to zoom, pan, and select specific periods of the analyzed time series.

Dhiria can assist clients in integrating these results into their business intelligence systems, CRMs, and other legacy systems within their ICT infrastructure, or support them in designing and developing a Business Intelligence system that integrates data from various corporate sources with Machine Learning results produced by Dhiria, allowing for improved business processes and decision-making.

D4TimeX is currently the most advanced system for time series exploration and prediction available in an as-a-service model.