Past Projects

Healthcare — NHS/Western General Hospital, Edinburgh

"Diagnostic of Abdominal Pain"

The Issues/Problems To improve medical diagnostics risk assessment of abdominal pain.
Project Activities Total of 9 (abdominal pain) diagnostic groups assessed; Number of attributes: 135 for abdominal pain;
Project Results Improved risk assessment for 9 abdominal pain from 65% of accuracy to over 76% with 90% level of confidence.

Healthcare — NHS/University College Hospital, London

"Proteomic Analysis of the Human Serum Proteome"

The Issues/Problems To diagnose ovarian, breast and heart diseases in large populations, using human serum proteome and to establish biomarkers in high-throughput protein analysis.
Project Activities Proteomics and data analysis of ovarian cancer, breast cancer and heart diseases have been performed; Number of attributes: 100,000 mass spectrometry peaks in the original data;
Project Results Our exploratory analysis shows that combining a single MS peak with CA125 allows statistically significant discrimination at the 5% level between cases and controls up to 14 months in advance of the original diagnosis of ovarian cancer. Such combinations work much better than a single peak or CA125 alone.

Environment — CIPEL

"Pollution Level Estimation"

The Issues/Problems To study of water and sediment pollution of Lake Geneva.
Project Activities From CIPEL (International Commission for the Leman Lake Water Protection), data was collected over a 10 year period between 1978 and 1988; The studies on the heavy metals contained in the sediments were already carried out; The studies of the space-time evolution of the chemical composition of water have started; In Lake Geneva, cadmium pollution becomes legally significant when it exceeds 0.8 g/l.
Project Results Produced a map of water and sediment pollution

Industry — Windfarm Generator Client

"Demand Optimisation"

The Issues/Problems A software company was developing a programme for a Spanish windfarm energy generator in order to optimise demands and generation outputs 8h, 4h and 1h in advance with 8 other parameters such as wind speed and direction taken into consideration.
Project Activities 9,000 training and 3000 testing examples in each of them; The attributes are three time series (8-14, 15-21, 23-29) and 8 additional attributes; The label to be predicted is energy T8, T4, T1 hours ago.
Project Results Improved predicted regression estimation method: mean absolute error from 1653 to 1252 (+1h) (24% improvement).

Industry — Navoi Mining and Metallurgical Company

"Estimation of expected gold concentration (grade) in the rocks" (This work was awarded by the State Prize of the USSR)

"Three-dimensional computer model of gold concentration over a deposit"

The Issues/Problems The project was aimed to develop an online system to estimate the gold grade distribution at one of the largest open pit gold mine at Muruntau (KyzylKum Desert, Uzbekistan). The Muruntau plant produces 1.7 million ounces of gold per year. The aim was to process 14 million metric tons per year of low grade stockpiles to produce 400,000 oz of gold per year.
Project Activities The main technical tool for implemented system was non-linear Kriging. This technique (similar to kernel ridge regression ) was used to estimate expected gold concentration (grade) in the rock surrounding each borehole. Anisotropy axes and corresponding correlation radiuses are automatically computed based on sampling data.
Project Results The system is in routine use at several gold deposits (including Muruntau) owned by Novoi Company and spread over KysylKum desert.