Akademik Veri Yönetim Sistemi
International Conference on Artificial Intelligence and Applied Mathematics in Engineering (ICAIAME 2022), Baku, Azerbaycan, 21 - 22 Mayıs 2022, ss.1-3
This study, which was conducted to facilitate the detection of neurodegenerative diseases such as Amyotrophic Lateral Sclerosis, Parkinson's disease and Huntington's disease, is a hybrid classification study that includes deep learning and machine learning methods. Amyotrophic lateral sclerosis (ALS) is a disease caused by the loss of motor nerve cells in the central nervous system, the spinal cord and the brain stem. Loss of these cells leads to muscle weakness and wasting (atrophy). Parkinson's disease (PD) is a nervous system disease due to a disorder of the gray matter nuclei in the lower parts of the brain. The basic disorder is in the parts of the brain that regulate coordinated movements. Huntington's disease (HD) is a genetic neurodegenerative brain disease. In addition to some movement disorders, patients have mental retardation. The most obvious symptoms are sudden and involuntary contractions of the arms, legs, trunk and face muscles. The dataset used to classify these diseases is public and has gait rhythm data from 13 ALS patients, 15 Parkinson's patients, 20 Huntington's patients and 16 healthy individuals (Control). Since the data from both legs are time series, LSTM architecture was chosen as the most suitable deep learning method. Using different machine learning methods (kNN, Random Forest, MLP, Bayes, Decision Tree, SVM, Gradient Boost Classifier, Logistic Regression) to create a hybrid method with LSTM, the method with the highest performance was determined as MLP. Various features (Maximum, minimum, Kurtosis, Skewness) were extracted from the data and tested with the hybrid LSTM-MLP method. The accuracy rate was 96.89% when diseased and healthy individuals were classified, the highest accuracy rate was 98.52% (HD vs. CO) when each disease and control group were classified, and 84.01% when all diseases were classified separately. As a result of experimental studies, it has been seen that the hybrid LSTM-MLP method designed outperforms the standard LSTM and standard MLP method. According to the obtained accuracy rates, the Hybrid LSTM-MLP method has been a system design that supports the decision mechanism of clinicians.