Comparative Analysis of Different Supervised Machine Learning Models for Recognizing Epilepsy in Children

Abstract

As we all know, Children are the future of every country, so it’s the vital fact that every country in this world wants to reduce the mortality rate of children in their country and improves their health also. By doing this, that particular country can enhance and develop in many ways like building unity, increase economic prosperity and political stability and many more. So, it’s a prime responsibility of every country in world wide web to take major steps for enhancing children health. Manual assessment of Epilepsy detection from electroencephalogram (EEG) waveform is very old-dated technique and mainly depends on the knowledge level of healthcare expert. Out of 100 cases, approximately 30 cases are wrongly interpreted due to the lack of automated approach and advancements in this field. The prediction methodology of EEG signals seems more powerful and accurate as scientists and researches moves towards the computational methods and techniques in order to take boom in healthcare industry. This paper presented a comparative analysis of several supervised machine learning models in order to recognize epilepsy in children using EEG data. Authors has taken the total 40 EEG signals of normal children and epileptic children. For generating statistical feature values (mean, median, mode and standard deviation) of experimental EEG signals, power spectral and fuzzy entropy techniques have been used. For classification part, five supervised machine learning algorithms (AdaBoost, support vector machine, naïve bayes, random forest and K nearest neighbor) were used. Experiments were carried out on the CHB-MIT scalp EEG database, channel FP1-F7 on 256 HZ data sampling rate. Results are generated by visualizing and evaluating the computed statistical EEG feature values using heat map and lift curve respectively. Performance evaluation has been performed on the basis of different classifiers used by other researchers. On the basis of classification results, authors conclude that 100% accuracy is achieved by analyzing different supervised machine learning models for detecting epilepsy in children.