When Machines can Understand: Seizure Prediction using Machine Learning

academicmemories
Aug 21
4 min read

Written by: Annette Dao

Edited by: Lea Kallo

The Human Mind

At what point does human cognition no longer become only understandable to us, but to the machines that we build? The intricacies of our mind separate us from the millions of species inhabiting our world. Quite simply, our brain is what makes us, you and me, an anomaly to intelligence. Does our consciousness define us as human or the centuries of behavior we evolved from? René Descartes' describes consciousness in a single statement, “I think, therefore I am.” Human behavior can be simplified to an even larger degree, to a single word: data.

Intelligence is a psychological phenomenon: hypothetical, abstract, and unobservable (Gignac and Szodorai, 2024). In our mortal fashion, we try to create what cannot be determined. Our artificial modelling of intelligence, commonly referenced as AI, attempts to comprehend the human mind in heuristics that is beyond mundane pattern recognition. AI knows us, learns us, and reads us through data. What happens when these machines try to predict us?

What is Epilepsy?

Epilepsy is a neurological disorder that causes recurring seizures in 1.2% of the United States’ population (“Epilepsy: Symptoms”, 2023). Characterized by an assortment of symptoms like muscle convulsions, loss of awareness, or abnormal brain activity, this disorder arises from abnormal electrical discharges in the brain (“Epilepsy: Symptoms”, 2023). Using electroencephalograms (EEG), electrode-based nodal discs in the scalp, we can interpret the mind’s activity as we see fit– especially in seizure monitoring (“EEG”, 2023). However, what humans cannot see, humans cannot deduce. When EEG signalling sources become too convoluted for human analysis, seizure prediction cannot occur. Thus, a different method for understanding epileptic episodes is needed.

Machine learning (ML) is focused on enabling computers to use behavior heuristics in imitating human performance (IBM, 2021). The intersection of human limitations to machine adaptability can create an ironic concept of machines understanding humans better than humans do.

How do Machines Interpret Human Behavior?

To formulate a thought, a connection, humans have a hierarchical structure of deduction that enables us to create a revelation. When we see an action or change in our surroundings, the information from our eyes travels to our mind and tells us what we do not directly observe, a pattern. In a similar manner, machine learning employs this architecture. To understand data is to categorize it, and in this swift action, pre-processing occurs. EEG signals from seizures are split into two classifications: the pre-ictal phase and interictal phase (Engel and Schwartzkroin, 2007). The pre-ictal phase is characterized by the aura, a subjective sensation that begins before an epileptic episode has occurred. On the other hand, the interictal phase is the period between seizures where there is a return to normal brain activity (Engel and Schwartzkroin, 2007). By sectioning the EEG signals into specific sections representing seizures, we can create a Fourier Transform analysis. (Zhu et al., 2024).

Fourier Transform models determine the sinusoidal frequency of EEG signals (Zhu et al., 2024) which reveal the different oscillations and periodicities that make up an original signal. Using this information, the model is converted into a spectrogram– showing changes to frequency content of a signal over time (Zhu et al., 2024).

Choosing the Brain: How do Machines “Think”?

After raw EEG signals have been preprocessed and segmented, a framework for neural connectivity is needed. Varying neural networks help extract patterns from data mirroring how humans deduce their line of thought. From the processed information, neural structures are used for deep feature extraction and classification of information. Convolutional neural networks (CNN) excel at using layers to detect different spatial hierarchies of features– making them adept at detecting morphologic signatures in EEG plots (“What Is a Convolutional”, n.d.). Conversely, recurrent neural networks (RNN) use sequential data or time-series dependent thinking to track how brain rhythms evolve over time (Stryker, 2024). Since spatial structure and temporal context are necessary in interpreting EEGs, a fusion model can combine CNN and RNN advantages and yields better results in seizure prediction (Zhang, 2024). More recently, transformer networks have advanced the field by reducing the recurrence with self-attention and combining the data processing with model training ideals. Together, these frameworks allow EEG signals to be used for clinical predictions.

The Consciousness of Thought

The ability for machines to interpret the human mind accurately should not be understated. We reach a point where human cognition no longer becomes only understandable to us, but to the technology that we build. This emerging field of machine learning and artificial intelligence opens doors to greater opportunities in neuroscience: biomarker discoveries, personalized therapies, and revealing brain dynamics that were once hidden from human analysis. We reached a point where human behavior no longer dictates our mortality. Instead, our consciousness is what makes us truly unique. At what point will machines learn “I think, therefore I am?”

References

Engel J.J., & Schwartzkroin P.A. (2007, May 9). What should be Modeled? Journal of Models of Seizures and Epilepsy, 1-14.

Gignac, G. E., & Szodorai, E. T. (2024). Defining intelligence: Bridging the gap between human and artificial perspectives. Intelligence, 104, Article 101832. https://doi.org/10.1016/j.intell.2024.101832 ScienceDirect+5EconPapers+5IDEAS/RePEc+5

IBM. (2021, September 22). What is machine learning? https://www.ibm.com/think/topics/machine-learning

MayoClinic. (2023, October 14). Epilepsy: Symptoms and causes. Mayo Clinic. Retrieved August 3, 2025, from https://www.mayoclinic.org/diseases-conditions/et pilepsy/symptoms-causes/syc-20350093

MayoClinic. (2023, December 8). EEG (electroencephalogram). Mayo Clinic. https://www.mayoclinic.org/tests-procedures/eeg/about/pac-20393875

Stryker, C. (2024, October 4). What is a recurrent neural network (RNN)? IBM Think. https://www.ibm.com/think/topics/recurrent-neural-networks

What Is a Convolutional Neural Network? (n.d.). MathWorks. Retrieved August 4, 2025, from https://www.mathworks.com/discovery/convolutional-neural-network.html

Zhang, X., Zhang, X., Huang, Q., & Chen, F. (2024). A review of epilepsy detection and prediction methods based on EEG signal processing and deep learning. Frontiers in Neuroscience, 18, Article 1468967. https://doi.org/10.3389/fnins.2024.1468967

Zhu, R., Pan, W.-x., Liu, J.-x., & Shang, J.-l. (2024). Epileptic seizure prediction via multidimensional transformer and recurrent neural network fusion. Journal of Translational Medicine, 22, Article 895.