AI Machine Learning

Machine Learning Prediction

Development of predictive models around structures, processes, components, experimental conditions and performance indicators, supported by interpretable results for screening, assessment and scientific decision-making.

Solution Overview

Establishment of a mechanical learning prediction process for scientific experiments, material performance, molecular properties, process parameters and test results, from pre-processing of data, feature engineering, modelling training, cross-certification to interpretation analysis. Projects could be used for small sample exploration, mass screening, risk identification and forecasting tool development.

Data preprocessingClean-up of deficiencies, anomalies, duplications and unit inconsistencies

Model developmentReturn, classification, sequencing and uncertainty assessment

Performance assessmentCross-certification, external validation and error analysis

Interpretation analysisDescription of key variables, sensitivities and scope of application

Applications

Material performance, reaction rate, adsorption capacity and stability projections
Modelling of the properties of drug molecules, catalysts, polymers and functional materials
Non-linear relationship analysis between experimental conditions and target indicators
Limited data available, prior feasibility assessment and model baseline required
Desiring to encapsulate the predictive model as an internal query tool or web application

Scientific data cleansing, quality control and profiling — Sample Cleaning

Scientific data cleansing and profiling — Data preprocessing

Machine learning model training and validation — Model training

Results of model interpretation and prediction assessment — Interpretation assessment

Overall process

1. Definition of objectivesProjection
Evaluation indicators

2. Data collationClean up the noise.
Feature Build

3. Model trainingAlgorithm Comparison
Parameter Optimization

4. Assessment of resultsError Analysis
Cross-validation

5. Interpretation outputContribution of variables
Scope of application

6. Deployment applicationsScript Tool
Web Interface

Technical modules

Feature Project:Explanatory features based on molecular, material, process or experimental conditions.
Model combination:Select linear models, tree models, nuclear methods, neural networks or integrated learning by data scale.
Certification system:Set up training/certification/test splits, cross-certification and external sample validation.
Uncertainty assessment:Output creditable interval, abnormal sample tips and scope description.

Project value

Rapid establishment of baselinesTo judge whether the data available are sufficient to support the prediction mission

Improved screening efficiencyBatch projection and prioritization of candidate samples

Risk reduction in decision-makingIdentification of high-risk results using error analysis and uncertainty

Facilitation of follow-up deploymentModels can be encapsulated as scripts, web pages or internal tools

Solution Overview

Applications

Overall process

Technical modules

Project value

More AI Services