A MapReduce-based rotation forest classifier for epileptic seizure prediction

Abstract—In this era, big data applications including biomedical are becoming attractive as the data generation and storage is increased in the last years. The big data processing to extract knowledge becomes challenging since the data mining techniques are not adapted to the new requirements. In this study, we analyse the EEG signals for epileptic seizure detection in the big data scenario using Rotation Forest classifier. Specifically, MSPCA is used for denoising, WPD is used for feature extraction and Rotation Forest is used for classification in a MapReduce framework to correctly predict the epileptic seizure. This paper presents a MapReduce-based distributed ensemble algorithm for epileptic seizure prediction and trains a Rotation Forest on each dataset in parallel using a cluster of computers. The results of MapReduce based Rotation Forest show that the proposed framework reduces the training time significantly while accomplishing a high level of performance in classifications. Keywords— Electroencephalogram (EEG); Epileptic Seizure prediction; Multi-scale Principal Component Analysis (MSPCA); Wavelet Packet Decomposition (WPD); Rotation Forest; Hadoop; Mapreduce.

1. INTRODUCTION Epilepsy is a neurological disorder characterized by a frequent tendency of the brain to yield abrupt bursts of abnormal electrical activity [1]. Such occurrences are called seizures and occur randomly. Excessive and synchronized activity of neurons causes epileptic seizures [2, 3]. Epilepsy is the second most common neurological disorder after strokes affecting over 1% of the world’s population [4]. In order to diagnose and identify the epileptic seizures, mostly the patient’s EEG must be monitored for several days. The process of monitoring is boring, time-consuming and expensive. Neurologist needs to determine the enduring epileptic activity from the recorded EEG data in order to determine if the used medication is working or not. Hence, a computer aided epileptic seizure detection system is extremely important [5, 6]. Electroencephalogram (EEG) has been used for clinical diagnosis of epilepsy for many decades. Compared to other methods such as Electrocorticogram (ECoG), EEG is a safe and clean method for detecting the activity of the brain. Clinical analysis of EEG traces for identification of seizures is well established. However, the performance of automated EEG based methods is dependent on the types of features analyzed and how they are used to classify the signal. Patients with epilepsy suffer from repeated seizures that manifest as physical or behavioral changes, which require intervention using medications or surgery [7]. A lot of channels are used for recording EEG signals. Processing of that number of channels is time consuming. Because of that, parallel processing is very important aspect of EEG signal processing in order to decrease processing time.
   Since EEG signals with many channels and complicated signal processing and classification algorithms cannot be analysed easily by means of personal computers. The cloud computing is the practical solution to these kind of big data problems [8]. Hadoop is considered to work on cloud build from thousands of commodity hardware nodes. Biomedical signal processing and classification algorithms can be parallelized to save computing time. Message passing interface (MPI) which is a traditional parallelization method may cause whole procedure to fail. Hadoop is a fault tolerant platform which makes it natural choice for these types of algorithms. Hadoop has lately been utilized in different areas of big data analysis [9]. Parallel computing is a concurrent computing which uses a group of autonomous processors employed together to solve a large computational problem. The essence of parallel computing is to partition and distribute the whole computational work among the processors [10]. Google’s MapReduce programming model [11] offers an effective structure for processing large datasets in a parallel manner. The Google File System [12] which inspires MapReduce delivers effective and consistent distributed data processing essential for applications including large datasets. The basic role of the MapReduce model is to support parallelism in which the programmer can benefit from the issues of distributed and parallel programming. Furthermore, MapReduce implementation includes load balancing, network performance, fault tolerance etc. [11]. The Apache Hadoop [13] is the most widely used ope

…(Full text truncated)…