Improving Neural Network Performance with Feature Selection Using Pearson Correlation Method for Diabetes Disease Detection

Diabetic or silent killer diseases are an alarming scourge for the world and classed as serious diseases. In Indonesia, increase in diabetics occurred by 2% vulnerable times between 2013 to 2018. This affects all sectors, both medical services financial sector. The Neural Network method a data mining algorithm is present overcome burden that arises early detection analysis of onset disease. However, has slow training capabilities can identify important attributes resulting decrease performance. Pearson correlation good at handling with mixed-type measuring information labels. With this, purpose this study will be use selection features improve neural network performance diabetes measure extent accuracy obtained from method. dataset used 130-US hospital UCI record number 101767 many 50 attributes. results found 94.93% 96.00%. As evaluation on AUC value increased 0.8077 0.8246. Thus Pearson's Correlation work well feature methods provide solutions improved accuracy.