Page 310 - Data Science Algorithms in a Week

Artificial Intelligence for the Modeling and Prediction ... 291

respective anti-inflammatory properties by a genetic algorithm. After further study using
a decision tree classifier, some 11 chemical compounds were determined to be
‘biomarkers’ of the putative anti-inflammatory potential. From these data, an
unsupervised model to predict new biologically active extracts from Asteraceae species
using HPLC-MS-MS information was built using an ANN with the back-propagation
algorithm using the biomarker data resulting in a high percentage of correct predictions
for dual inhibition.
Nagahama et al. (2011) proposed the simultaneous estimation of the multiple health-
promoting effects of food constituents using ANNs. The model utilizes expression data of
intracellular marker proteins as descriptors that reply to stimulation of a constituent. To
estimate three health-promoting effects, namely, cancer cell growth suppression activity,
antiviral activity, and antioxidant stress activity, each model was constructed using
expression data of marker proteins as input data and health-promoting effects as the
output value.
Goodacre et al. (1998) used unsupervised methods of discriminant function and
hierarchical cluster analyses to group the spectral fingerprints, of clinical bacterial
isolates associated with urinary tract infection. ANNs trained with Raman spectra
correctly identified some 80% of the same test set, thus providing rapid accurate
microbial characterization techniques, but only when combined with appropriate
chemometrics.
Zeraatpishe et al. (2011) studied the effects of Lemon balm infusions (30 days, twice
daily, a tea bag of 1.5 g in 100 mL water) on the oxidative stress status in radiology staff
exposed to persistent low-dose radiation during work. They measured lipid peroxidation,
DNA damage, catalase, superoxide dismutase, myeloperoxidase, and glutathione
peroxidase activity in plasma samples. The treatment markedly improved oxidative stress
condition and DNA damage in radiology staff. The authors posed the question whether
our approach to apply ANNs to correlate with the antioxidant essential oils (Cortes-
Cabrera & Prieto, 2010) was to be applied to the protective activities of Lemon balm in
order to improve this intervention.

FACTORS INFLUENCING THE ACCURACY OF

THE PREDICTIONS: STRATEGIES TO MINIMISE THEM

Internal Factors

Some of the reported problems in the application of ANNs are caused by their
inherent structure and the most important are ‘overtraining’, ‘peaking effect’, and
‘network paralysis’. Overtraining the ANN may lead to the noise of data used for training