Detecting malicious software using sensors
In some implementations, a method includes retrieving data from multiple sensors in a computing device, and the multiple sensors comprise different types of sensors. The sensor data is analyzed based on a predictive model, and the predictive model is trained to detect malware. Initiation of malware is determined based on the analysis. In response to the determination, the malware is terminated.
1. A non-transitory computer readable medium storing instructions to cause a processor of a computing device to perform operations comprising:
simulating initiation of a ransomware encryption in the computing device;
while the ransomware encryption is executed by the computing device, obtaining sensor data from multiple sensors in the computing device, wherein the multiple sensors comprise different types of sensors to monitor an operating condition of internal hardware components of the computing device, the multiple sensors residing in a side channel separate from the processor of the computing device, and the obtained sensor data reflective of the operating condition of the internal hardware components while the ransomware encryption is executed by the computing device, wherein the obtained sensor data obtained while the ransomware encryption is executed by the computing device is different from sensor data obtained while the computing device is not executing the ransomware encryption;
training a predictive model to detect malware using the obtained sensor data to detect execution of ransomware by the computing device;
after training the predictive model, obtaining sensor data from multiple sensors in the computing device during a normal operation of the computing device;
analyzing the obtained sensor data retrieved during the normal operation based on the predictive model;
determining initiation of malware in response to analyzing the obtained sensor data retrieved during the normal operation based on the predictive model; and
in response to the determination, terminating the malware.
2. The non-transitory computer readable medium of claim 1 , wherein the malware comprises ransomware, and initiation of ransomware encryption is detected.
3. The non-transitory computer readable medium of claim 1 , wherein the predictive model comprises a feature vector determined using machine learning.
4. The non-transitory computer readable medium of claim 1 , wherein the multiple sensors comprise a sensor for at least one of a main memory power, voltage, current, or temperature.