Deteksi indikasi penyakit asma menggunakan Multi Layer Perceptron (MLP) berdasarkan data rekam medis

INDONESIA:

Penyakit asma merupakan salah satu penyakit pernapasan kronis yang dapat menurunkan kualitas hidup penderitanya sehingga diperlukan deteksi dini untuk membantu penanganan yang lebih cepat dan tepat. Penelitian ini bertujuan untuk menganalisis performa metode Multilayer Perceptron (MLP) dalam mendeteksi indikasi penyakit asma berdasarkan data rekam medis serta mengevaluasi pengaruh penyeimbangan data menggunakan Synthetic Minority Oversampling Technique (SMOTE) dan variasi arsitektur jaringan terhadap kinerja model. Penelitian dilakukan melalui tahapan pra-pemrosesan data, penyeimbangan kelas menggunakan SMOTE, perancangan beberapa arsitektur MLP, serta evaluasi menggunakan metode 5-Fold Cross Validation. Hasil pengujian pada kondisi data imbalanced menghasilkan rata-rata Accuracy sebesar 97,95%, Precision sebesar 85,64%, Recall sebesar 73,46%, dan F1-Score sebesar 78,65%. Sementara itu, pada kondisi data balanced menggunakan SMOTE diperoleh rata-rata Accuracy sebesar 97,94%, Precision sebesar 83,84%, Recall sebesar 74,95%, dan F1-Score sebesar 78,92%. Konfigurasi arsitektur terbaik diperoleh pada model 20-64-32-16-1 dengan data balanced yang menghasilkan Accuracy sebesar 98,12%, Precision sebesar 84,38%, Recall sebesar 78,74%, dan F1-Score sebesar 81,22%. Hasil penelitian menunjukkan bahwa penerapan SMOTE mampu meningkatkan kemampuan model dalam mengenali kelas minoritas, sedangkan peningkatan jumlah neuron memberikan pengaruh yang lebih konsisten terhadap performa model dibandingkan penambahan hidden layer. Secara keseluruhan, metode Multilayer Perceptron (MLP) dapat digunakan secara efektif untuk mendeteksi indikasi penyakit asma berdasarkan data rekam medis dengan performa klasifikasi yang baik.

ENGLISH:

Asthma is a chronic respiratory disease that can reduce the quality of life of sufferers, necessitating early detection to facilitate faster and more appropriate management. This study aims to analyze the performance of the Multilayer Perceptron (MLP) method in detecting indications of asthma based on medical record data, as well as to evaluate the effect of data balancing using the Synthetic Minority Oversampling Technique (SMOTE) and variations in network architecture on model performance. The research was conducted through the stages of data preprocessing, class balancing using SMOTE, designing several MLP architectures, and evaluation using the 5-Fold Cross-Validation method. Testing on the imbalanced data condition yielded an average Accuracy of 97.95%, Precision of 85.64%, Recall of 73.46%, and an F1-Score of 78.65%. Meanwhile, under the balanced data condition using SMOTE, the average Accuracy was 97.94%, Precision was 83.84%, Recall was 74.95%, and the F1-Score was 78.92%. The best architectural configuration was achieved by the 20-64-32-16-1 model with balanced data, producing an Accuracy of 98.12%, Precision of 84.38%, Recall of 78.74%, and an F1-Score of 81.22%. The results demonstrate that the application of SMOTE enhances the model's ability to recognize the minority class, while increasing the number of neurons provides a more consistent impact on model performance compared to adding hidden layers. Overall, the Multilayer Perceptron (MLP) method can be effectively utilized to detect indications of asthma based on medical record data with strong classification performance.

ARABIC:

ﻳﻌُﺪ ﻣﺮﺽ ﺍﻟﺮﺑﻮ ﺃﺣﺪ ﺃﻣﺮﺍﺽ ﺍﻟﺠﻬﺎﺯ ﺍﻟﺘﻨﻔﺴﻲ ﺍﻟﻤﺰﻣﻨﺔ ﺍﻟﺘﻲ ﻳﻤﻜﻦ ﺃﻥ ﺗﻘﻠﻞ ﻣﻦ ﺟﻮﺩﺓ ﺣﻴﺎﺓ ﺍﻟﻤﺼﺎﺑﻴﻦ ﺑﻪ، ﻟﺬﺍ ﻓﺈﻥ ﺍﻟﻜﺸﻒ ﺍﻟﻤﺒﻜﺮ ﺿﺮﻭﺭﻱ ﻟﻠﻤﺴﺎﻋﺪﺓ ﻓﻲ ﺍﻟﻌﻼﺝ ﺑﺸﻜﻞ ﺃﺳﺮﻉ ﻭﺃﻛﺜﺮ ﺩﻗﺔ. ﺗﻬﺪﻑ ﻫﺬﻩ ﺍﻟﺪﺭﺍﺳﺔ ﺇﻟﻰ ﺗﺤﻠﻴﻞ ﺃﺩﺍﺀ ﻃﺮﻳﻘﺔ ﺍﻟﺸﺒﻜﺔ ﺍﻟﻌﺼﺒﻴﺔ ﻣﺘﻌﺪﺩﺓ ﺍﻟﻄﺒﻘﺎﺕ (MLP) ﻓﻲ ﺍﻟﻜﺸﻒ ﻋﻦ ﻣﺆﺷﺮﺍﺕ ﻣﺮﺽ ﺍﻟﺮﺑﻮ ﺍﺳﺘﻨﺎﺩﺍً ﺇﻟﻰ ﺑﻴﺎﻧﺎﺕ ﺍﻟﺴﺠﻼﺕ ﺍﻟﻄﺒﻴﺔ، ﻭﺗﻘﻴﻴﻢ ﺗﺄﺛﻴﺮ ﻣﻮﺍﺯﻧﺔ ﺍﻟﺒﻴﺎﻧﺎﺕ ﺑﺎﺳﺘﺨﺪﺍﻡ ﺗﻘﻨﻴﺔ ﺃﺧﺬ ﺍﻟﻌﻴﻨﺎﺕ ﺍﻟﺰﺍﺋﺪﺓ ﻟﻸﻗﻠﻴﺔ ﺍﻻﺻﻄﻨﺎﻋﻴﺔ (SMOTE) ﻭﺗﺄﺛﻴﺮ ﺗﻨﻮﻉ ﺑﻨﻴﺔ ﺍﻟﺸﺒﻜﺔ ﻋﻠﻰ ﺃﺩﺍﺀ ﺍﻟﻨﻤﻮﺫﺝ. ﺃﺟُﺮﻳﺖ ﺍﻟﺪﺭﺍﺳﺔ ﻣﻦ ﺧﻼﻝ ﻣﺮﺍﺣﻞ ﻣﺎ ﻗﺒﻞ ﻣﻌﺎﻟﺠﺔ ﺍﻟﺒﻴﺎﻧﺎﺕ، ﻭﻣﻮﺍﺯﻧﺔ ﺍﻟﻔﺌﺎﺕ ﺑﺎﺳﺘﺨﺪﺍﻡ ﺗﻘﻨﻴﺔ SMOTE، ﻭﺗﺼﻤﻴﻢ ﻋﺪﺓ ﻫﻴﺎﻛﻞ ﻟﺸﺒﻜﺔ MLP، ﻭﺍﻟﺘﻘﻴﻴﻢ ﺑﺎﺳﺘﺨﺪﺍﻡ ﻃﺮﻳﻘﺔ ﺍﻟﺘﺤﻘﻖ ﺍﻟﻤﺘﻘﺎﻃﻊ ﺍﻟﺨﻤﺎﺳﻲ Cross-5) Fold .(Validation ﺃﺳﻔﺮﺕ ﻧﺘﺎﺋﺞ ﺍﻻﺧﺘﺒﺎﺭ ﻓﻲ ﺣﺎﻟﺔ ﺍﻟﺒﻴﺎﻧﺎﺕ ﻏﻴﺮ ﺍﻟﻤﺘﻮﺍﺯﻧﺔ ﻋﻦ ﻣﺘﻮﺳﻂ ﺩﻗﺔ (Accuracy) ﺑﻠﻎ %97,95، ﻭﺩﻗﺔ ﺗﺤﺪﻳﺪ (Precision) ﺑﻠﻐﺖ %85,64، ﻭﻣﻌﺪﻝ ﺍﻻﺳﺘﺮﺟﺎﻉ (Recall) ﺑﻠﻎ %73,46، ﻭﻣﺆﺷﺮ F1-Score ﺑﻠﻎ .%78,65 ﻓﻲ ﺍﻟﻤﻘﺎﺑﻞ، ﻓﻲ ﺣﺎﻟﺔ ﺍﻟﺒﻴﺎﻧﺎﺕ ﺍﻟﻤﺘﻮﺍﺯﻧﺔ ﺑﺎﺳﺘﺨﺪﺍﻡ SMOTE، ﺗﻢ ﺍﻟﺤﺼﻮﻝ ﻋﻠﻰ ﺩﻗﺔ ﻣﺘﻮﺳﻄﺔ ﺑﻠﻐﺖ %97,94، ﻭﺩﻗﺔ ﺗﺤﺪﻳﺪ ﺑﻠﻐﺖ %83,84، ﻭﻣﻌﺪﻝ ﺍﺳﺘﺮﺟﺎﻉ ﺑﻠﻎ %74,95، ﻭﺩﺭﺟﺔ F1 ﺑﻠﻐﺖ .%78,92 ﺗﻢ ﺍﻟﺤﺼﻮﻝ ﻋﻠﻰ ﺃﻓﻀﻞ ﺗﻜﻮﻳﻦ ﻟﻠﻬﻨﺪﺳﺔ ﺍﻟﻤﻌﻤﺎﺭﻳﺔ ﻓﻲ ﺍﻟﻨﻤﻮﺫﺝ 1-16-32-64-20 ﻣﻊ ﺍﻟﺒﻴﺎﻧﺎﺕ ﺍﻟﻤﺘﻮﺍﺯﻧﺔ، ﻣﻤﺎ ﺃﺩﻯ ﺇﻟﻰ ﺩﻗﺔ ﺑﻨﺴﺒﺔ %98,12، ﻭﺩﻗﺔ ﺗﺤﺪﻳﺪ ﺑﻨﺴﺒﺔ %84,38، ﻭﻣﻌﺪﻝ ﺍﺳﺘﺮﺟﺎﻉ ﺑﻨﺴﺒﺔ %78,74، ﻭﺩﺭﺟﺔ F1 ﺑﻨﺴﺒﺔ .%81,22 ﺃﻇﻬﺮﺕ ﻧﺘﺎﺋﺞ ﺍﻟﺒﺤﺚ ﺃﻥ ﺗﻄﺒﻴﻖ SMOTE ﻗﺎﺩﺭ ﻋﻠﻰ ﺗﺤﺴﻴﻦ ﻗﺪﺭﺓ ﺍﻟﻨﻤﻮﺫﺝ ﻋﻠﻰ ﺍﻟﺘﻌﺮﻑ ﻋﻠﻰ ﺍﻟﻔﺌﺎﺕ ﺍﻷﻗﻠﻴﺔ، ﻓﻲ ﺣﻴﻦ ﺃﻥ ﺯﻳﺎﺩﺓ ﻋﺪﺩ ﺍﻟﺨﻼﻳﺎ ﺍﻟﻌﺼﺒﻴﺔ ﺗﺆﺛﺮ ﺑﺸﻜﻞ ﺃﻛﺜﺮ ﺍﺗﺴﺎﻗﺎً ﻋﻠﻰ ﺃﺩﺍﺀ ﺍﻟﻨﻤﻮﺫﺝ ﻣﻘﺎﺭﻧﺔً ﺑﺈﺿﺎﻓﺔ ﻃﺒﻘﺔ ﺧﻔﻴﺔ. ﻭﺑﺸﻜﻞ ﻋﺎﻡ، ﻳﻤﻜﻦ ﺍﺳﺘﺨﺪﺍﻡ ﻃﺮﻳﻘﺔ )MLP Perceptron (Multilayer ﺑﻔﻌﺎﻟﻴﺔ ﻟﻠﻜﺸﻒ ﻋﻦ ﻣﺆﺷﺮﺍﺕ ﻣﺮﺽ ﺍﻟﺮﺑﻮ ﺍﺳﺘﻨﺎﺩﺍً ﺇﻟﻰ ﺑﻴﺎﻧﺎﺕ ﺍﻟﺴﺠﻼﺕ ﺍﻟﻄﺒﻴﺔ ﻣﻊ ﺃﺩﺍﺀ ﺗﺼﻨﻴﻒ ﺟﻴﺪ