Predictive models

Machine Learning



Goal definition


Data collection

Data quality and quantity determine the model’s quality

Data preprocessing

Prepare and clean data


Choose and implement a model based on:

  • Data type
  • Goal


Train the model with your data to imporve its ability to predict


Test the model and evaluate the model performance


We define the right goal with your domain experts and make predictions or decision about it, using machine learning algorithms which learn from your data.

After having defined your objective with domain experts, we help you predicting outputs based on input data.  The models we choose and implement take into account your data type and goal.

Consumer Complains prediction

Using deep learning, we can predict consumer complains generated by a product based on multi-sectoral data (such as raw material, production line and quality control data).

The model prediction allows to reject products, not detected by quality control standard tests, that will generate high consumers complains.

Its analysis highlights the critical factors to monitor, to ensure better quality in production.

Number of emergency arrivals prediction

Based on a Deep Learning model, more precisely the combination of a multi-layered neural network composed of LSTM (Long-Short term memory) nodes and convolution operations, we offer to hospitals and specifically to their emergency department, a decision support through a prediction of the number of patient arrivals.

This tool extracts all the important information affecting the predictions while preserving the intrinsic dependencies of the sequential aspect of the available data.

In terms of results, this model allows us to reach an accuracy of 93.35%.

Anticipate future results

Support in decision making

Provide informing conclusions

Our partners



Research & Education

Contact us

+41 21 353 91 00

ProcSim, EPFL Innovation Park, Building D

CH – 1015 Lausanne