Scientists from Russia and other countries have developed a new algorithm that employs a neural network to identify the degree of disorder in data and assist in data classification using machine learning. Their research, published in Algorithms and supported by the Russian Scientific Fund, demonstrated the accuracy of the algorithm’s results after applying it to electroencephalograms of healthy people and patients with Alzheimer’s disease, with an accuracy of over 70%.
Traditionally, the concept of entropy is used to gauge the degree of chaos, disorder, and uncertainty in data from various fields of science and practice. The more turbulent the data, the harder it is to predict the activity starting. However, accurately calculating entropy is often challenging and susceptible to distortion. As a result, the scientists developed a unique method that uses a neural network to discern a unique type of entropy: NNETEN entropy (Neural Network Entropy – Entropy on the Neuron network).
Scientists used the MNIST database, a set of handwritten digits ranging from 0 to 9, to educate the NNETEN entropy neural network in time series, number sequences, or random values that vary over time. They subsequently assessed the algorithm’s efficiency with electroencephalograms in the study.
EEGs, which detect electrical activity in the brain, can assist in diagnosing various illnesses. One such disease that doctors can detect using EEGs is Alzheimer’s disease, which leads to memory loss and cognitive decline. The neural network was tasked with dividing a database of 65 patients with 36 of the patients having Alzheimer’s and 29 healthy patients for classification purposes using nneten entropy.
As the accuracy of NNETEN entropy alone was insufficient, scientists employed a variety of entropy types in concert with NNETEN entropy to optimise the algorithm. This combination of entropy types, such as approximate, permanent, or fuszi entropy, produced more significant classification accuracy, up to 73%, honing the algorithm’s potential to assist in recognising early signs of Alzheimer’s disease.
Researchers have found that this methodology can evaluate other types of data structures such as audio signals, seismic fluctuations, cardiograms, and currency pairs. The public domain contains the neural network used in calculating Nneten entropy, allowing other researchers to use it for their projects.