A Novel Mechanism for Tuning Neural Network for Malware Detection in Android Device

Malicious software or code that is specifically targeted towards Android devices, such as smartphones and tablets running the Android operating system, is known as Android malware. The objectives of these malicious programs can vary, but they all generally try to compromise the security and privacy of the device or its user. Detecting Android malware using machine learning is a challenging but effective approach, as it can help identify malicious apps based on patterns, behaviors, and features. This paper proposes a Tuned Neural Network model for the detection of Malware in Android. The parameters of the NN are tuned using hyperparameter tuning with Random Search. The principal component Analysis method for correlation detection, has been included such that the dimensionality reduction helps in faster execution. The Android Malware dataset's split between harmless and dangerous applications is frequently skewed, with a much greater proportion of benign instances. This discrepancy might result in skewed models. Are better at detecting benign apps than malware. Machine learning models rely on historical data, making them less effective at detecting new, previously unseen threats (zero-day vulnerabilities). Neural networks can be updated with new data and retrained periodically, allowing the detection system to adapt to emerging threats and zero-day vulnerabilities. It is critical to find a balance between accuracy (minimising false positives) & recall (minimising false negatives) in Android malware detection. Hyperparameter tuning can help adjust the model's threshold or other settings to achieve the desired trade-off between these two metrics. In the proposed model, it has acquired + 1.06% accuracy than traditional approaches and made the loss approximately equal to 0%.