Highly Intelligible Models
XTRACTIS-induced models are intelligible: humans can fully understand the internal decision-making logic of the AI systems automating the decisions.
This chart shows the “#1 ranking” counts for INTELLIGIBILITY observed from 24 Public Use Cases with benchmarks.
Highly Predictive Models
XTRACTIS-induced models are robust: their performance is at least as high as mainstream AI and data-driven modeling techs, as benchmarks prove.
This chart shows the “#1 ranking” counts for PERFORMANCE observed from 24 Public Use Cases with benchmarks.
As Logistic Regression is inapplicable to Regression (UC#02, UC#15 and UC#23) and unavailable for UC#01, graph#1 displays the scores calculated for the 27 UCs, but only for 4 modeling techniques. Graph#2 displays the scores of the 5 modeling techniques, calculated for the 23 UCs for which Logistic Regression is available.
All quantitative metrics are defined in the use cases documents.
The benchmarks of XTRACTIS, Logistic Regression, Random Forest, Boosted Tree, and Neural Network are carried out on Test and External Test datasets (all the unknown cases of the reference dataset, i.e., not used for training or validation).
In these graphs, the Intelligibility Score IS and the Performance Score PS are calculated from all the public Use Cases (UC) that include benchmarks of models.
The center of a bubble corresponds to the point of coordinates IS -on top- and PS -on bottom. A bubble on the top-right corner is the Holy Grail for critical AI-based decision systems: an AI Technique which produces predictive models with the highest Performance and the highest Intelligibility.
Our Use Cases present the results of XTRACTIS modeling and include complete Benchmarks against Neural Network, Boosted Tree, Random Forest, and Logistic Regression.
These studies illustrate the ability of XTRACTIS to automatically induce knowledge in the form of predictive and intelligible mathematical relationships from real-world data (public data or authorized private data).
For each application, we show how the induced decision system uses its fuzzy rules to compute explained predictions for new situations, i.e. unknown to the learning data set.
Choose the sector to access the corresponding Use Cases.
XTRACTIS discovers how to detect a diabetic condition oppotunistically from the patient’s blood test and demographic characteristics. The result is an explainable and accurate automated medical diagnosis.
XTRACTIS discovers how to detect the epileptic seizure from the patient’s electroencephalogram (EEG) processed signal. The result is an explainable and accurate automated medical diagnosis.
XTRACTIS discovers how to identify foetus heart conditions based on signal characteristics of fetal heart rate and the mother’s uterine contractions. The result is an explainable and accurate automated medical diagnosis.
XTRACTIS discovers how to detect breast cancer based on topological characteristics of mammary cells. The result is an explainable and accurate automated medical diagnosis.
XTRACTIS discovers how to detect ovarian cancer based on the 15,154 mass-charge ratios coming from the protein spectrum of serum samples, and for a small number of patients. The result is an explainable and accurate automated medical diagnosis.
XTRACTIS discovers how to detect prostate cancer based on the expression levels of 12,600 genes and for a small number of patients. The result is an explainable and accurate automated medical diagnosis.
XTRACTIS discovers how to identify 2 types of lung cancer based on the expression levels of 12,533 genes and for a small number of patients. The result is an explainable and accurate automated medical diagnosis.
XTRACTIS discovers how to detect kidney disease based on the patient record and blood measures. The result is an explainable and accurate automated medical diagnosis.
XTRACTIS discovers how to detect Parkinson’s disease based on simple voice recordings of patients. The result is an explainable and accurate automated medical diagnosis.
XTRACTIS discovers how we can automatically idenify and predict the wine’s origin from its physicochemical features in order to help preserve the wine’s protected designation of origin.
XTRACTIS discovers a company’s strategies to evaluate employees, highlighting the existing gender and age biases in these strategies.
XTRACTIS discovers a company’s strategies to evaluate employees, highlighting the existing ethnic bias in these strategies.
XTRACTIS robots explore a 27-dimension decision space of possible supply chain configurations to find the one that maximizes at best Operating Income given requests on local sales objectives, local production constraints, and ecological requirements. Objectives and constraints are either fuzzy (flexible) or binary (rigid).
Design an AI-based decision-making system that accurately and rationally identifies underwater sounds from their signal characteristics.
Design an AI-based decision system that accurately identifies risky behavior linked to criminal activities by analyzing communication metadata from surveillance investigations, without accessing the content of telephone calls and rationally predicts dangerous Homeland Security situations.
Design an AI-based decision system that accurately and instantly diagnoses an intrusion on a computer network from features of the connection logs, to execute the appropriate action based on a rational decision.
XTRACTIS makes the Reverse-engineering of a Boosted Tree model to transform this opaque model into a transparent rule-based decision system that accurately makes rational and explainable predictions.
XTRACTIS robots explore a 27-dimension decision space of possible supply chain configurations to find the one that maximizes at best Operating Income given requests on local sales objectives, local production constraints, and ecological requirements. Objectives and constraints are either fuzzy (flexible) or binary (rigid).
