Transfer Learning with Hybrid Firefly Butterfly Optimization Feature Selection Model for Early Parkinson Disease Prediction

Autor: Pal, Rekha, Pandey, Mithilesh Kumar, Pal, Saurabh
Zdroj: Biomedical Materials & Devices; 20240101, Issue: Preprints p1-12, 12p
Abstrakt: Parkinson’s disease classification is required to assist doctors for deciding diagnostic treatment. Many researchers conducted experiments to detect Parkinson disease in earlier stage but there exist several limitations like over-fitting problem, imbalanced data problem, and irrelevant features in the dataset. To deal with these problems, this paper proposes an efficient Parkinson disease classification model named Transfer Learning- Long Short-Term Memory. Transfer learning is applied with Long Short-Term Memory to improve performance of Parkinson disease classification. Transfer learning method selects the optimal batch size for the LSTM model to eliminate over-fitting and imbalanced problems. The most significant features in the dataset are selected using Hybrid Firefly Butterfly Optimization algorithm (HFBOA). Hybrid algorithm has the advantage of easily escape from local Optima and good conversions rate. The dataset used in this study contains 23 attributes and 197 instances are taken from University of California Irvine machine learning repository. This dataset analyzed the instances of 31 patients in which 23 patients were found as infected from Parkinson's disease and 8 patients are sound. The efficiency of the proposed TL-LSTM model is examined using different evaluate measures namely accuracy, precision, recall, and f1-score. The results of the study show that the TL-LSTM has achieved highest accuracy 98.90% when we apply HFBOA features selection algorithm. HFBOA selects 10 most significant features using 200 iterations with swarm size of 50. The results of experimental analysis display superior performance of the proposed TL-LSTM model with HFBOA feature selection over other existing methods particularly with the accuracy of 98.90%. The evaluation results display that the proposed novel approaches successfully identify and classify Parkinson disease problems.
Databáze: Supplemental Index
Externí odkaz: