Table 16 Ablation study on CICMalDroid 2020 dataset using HDDM–FedSGD configuration.

From: FL-MalDrift: a federated learning framework for malware detection under local concept drift

Configuration

Accuracy (%)

F1

Precision

Recall

Full FL-MalDrift (All Modules)

96.8

0.962

0.969

0.956

Without Drift Detection

92.1

0.917

0.921

0.914

Without Threshold Adaptation

93.4

0.929

0.936

0.922

Without Adaptive Aggregation

94.6

0.943

0.948

0.939

  1. Bold values indicate the best performance scores among the compared methods.
Back to article page