Time series analysis and predictive models

Time series data is a ubiquitous type of data, but its processing and analysis pose unique challenges. Handling this type of data correctly enables companies to predict future developments and thus gives them significant competitive advantages – and inovex can support you in this.

Home / Our services / Data Science / Time series analysis and predictive models

Overview & Offering

Time Series

Time series data is data which collects time-related information. This data is usually more complex than non-time-based data, and it therefore requires special handling. Our experts have advanced knowledge in analysing time series data, and we can help you gain valuable insights from yours.

We can help you with:

Exploration

Do you have time series data (data in chronological order), but you don’t know what added value it can furnish?

We’ll help you future-proof your ideas through:

Feasibility studies
Product discovery
Proofs of concept
Integration into your data and AI strategy

Implementation

Would you like to implement a specific business case using time series data?

We can provide you with end-to-end support for:

Data acquisition – from the sensor to the time series database
Analysing/ensuring data quality and temporal synchronisation
Data analysis and modelling, including:
- Pattern and trend recognition
- Anomaly detection
- Predictions
Visualisation, monitoring, reporting, and data storytelling

Background: Time series data is ubiquitous

Sensor data which measures the temperature of a machine component every few seconds, hourly weather data, daily patient data, monthly sales figures… These are all types of time series data, as they show developments and changes over time. Modern technology has enabled this type of valuable data to be generated on a massive scale. Time series data plays a role in every industry, and we encounter it in almost all of our customer projects:

Automotive: vehicle sales figures, development of fuel prices, production figures, vehicle maintenance intervals
Retail: product sales figures, inventory levels over time, customer behaviour patterns, shopping trends, seasonal demand for products, evaluation of advertising campaigns
Logistics: data from supply chains for raw materials and semi-finished and finished products, as well as product storage and shipping data
Financial sector: share prices, foreign exchange rates, raw materials prices, interest rates, inflation rates, stock market indices, bond yields, development of risk premiums
Healthcare: patient monitoring, hospital occupancy rates, infection rates, data from wearables, such as pulse, sleep, and stress indices
Food industry: production volumes, inventory quantities, expiry dates, prices, product quality
Media & Entertainment: TV ratings, online video views, social media activities, visitor numbers and website behaviour, concert attendance figures
Industry: production volumes; machine, sensor, and IoT data, maintenance histories, energy consumption patterns, quality control metrics

Case Studies

KROHNE Messtechnik GmbH: Real-time detection of anomalies in measuring instruments

As part of their Service Meister (Service Master) research project, SMEs KROHNE Messtechnik GmbH and inovex have designed a system which uses machine learning to automatically detect and evaluate anomalies in measuring instruments.

READ CASE STUDY

RATIONAL AG: ConnectedCooking – IoT Platform

With ConnectedCooking, inovex and RATIONAL have built an innovative IoT platform for the catering kitchen and rolled it out globally.

READ CASE STUDY

REWE digital: Demand Forecasting for REWE’s Delivery Service

inovex and REWE’s collaboration in the area of supply chain optimisation has focused particularly intensively on REWE’s IT subsidiary, REWE digital.
This particular project involved developing demand forecasting for REWE’s delivery service, leveraging big data technologies to enable easy scalability.

READ CASE STUDY

From patterns and trends to predictions

Time series analysis is an important method for understanding and analysing changes in data. It is used to identify patterns and trends in order to determine the factors underlying changes and (ultimately) to make predictions about future developments.

Adding time increases complexity

Adding the temporal dimension to data adds complexity and creates a significant challenge. Traditional statistical methods, for example, often reach their limits when it comes to time series data, which depicts changes and trends over time.

Not only does time series data provide information about individual observations, but it also reveals relationships between the successive data points. For example, the future temperature in London on a particular day is influenced not only by the current temperature in the region, but also by the city’s temperature the previous day. While this knowledge opens up a wide range of possibilities, it also requires a thorough understanding of the temporal component and the underlying mathematics.

Looking into the future

Determining the best algorithm for solving a particular problem depends on the domain-specific characteristics of the data and the goals of the analysis. Our customers benefit from our many years of experience in handling time series data from numerous application areas. We can support you in finding the right method for your use case:

Classical time series analysis uses established methods. These are easy to implement, usually require less computing power, and provide initial results quickly. They cannot, however, always fully grasp the complexity of data generated today, especially if it is non-stationary, non-linear, or contains outliers. Such cases require additional methods of analysis.

Mathematical modelling makes it possible to simulate different scenarios and to study their effects on the time series in order to identify potential risks and opportunities. It is particularly complex and requires extensive computing resources, as well as special expertise.

Machine learning (ML) is a popular tool for time series predictions. ML algorithms are particularly useful for complex data which cannot be adequately described using simple statistical models. Such data requires targeted preprocessing and feature extraction.

The application of deep learning eliminates the need for targeted feature extraction of time series data. Deep neural networks enable complex temporal patterns and relationships to be recognised almost automatically. Deep learning is particularly useful for complex data, but it requires more data and computational resources, and interpreting the results comes with its own set of challenges.

Hybrid methods for time series forecasting combine statistical and machine learning methods to achieve better results. Combining the strengths of both approaches enables the methods’ respective limitations to be overcome. A priori knowledge, for example, can help to considerably simplify the prediction tasks involved in machine learning methods, thus reducing the computational effort.

inovex as a partner

We will work with you to develop a comprehensive, customised concept for extracting the maximum level of information from your time series data and creating added value. Our experts in data engineering, data science, and machine learning will help you choose the right method, assist you with technical integration, and exemplify an agile mindset that will make your data product a success.