Researchers from the National Institute of Technology in Tiruchirappalli, Tamil Nadu, India, are using emotional data in combination with machine learning and deep learning techniques to create technology that promises to help better explain the criminal mind and potentially make predictions about criminal activity in a bid to possibly prevent it.

To accomplish these goals, the team analyzed voice-based emotional cues by using machine learning algorithms, which the team suggested led to a detection accuracy of 97.2% for various crimes.

Further, deep learning approaches — specifically, convolutional stacked bidirectional long short-term memory (LSTM) — enabled the research team to identify crime hotspots with a 95.64% rate of accuracy.

The researchers suggested that emotional states in speech patterns enabled them to examine speech-based emotion detection, taking linguistic origin, paralinguistic cues and the characteristics of the speaker into account. This reportedly enabled the team to incorporate the emotional data they acquired with other details including the location and type of crime taking place in a particular hotspot.

This technology could eventually be used to monitor activities in crime hotspots, detect crimes and make predictions about future criminal activities.

Further, the team suggested that such machine learning techniques might be used for future emergency response systems, in addition to crime fighting applications. For instance, the technology might be used to analyze the emotional content of a person calling emergency services and distinguish between genuine emergencies and non-emergency or even fraudulent calls, thereby reducing the burden on these services.

An article detailing the findings, “Crime detection and crime hot spot prediction using the BI-LSTM deep learning model,” appears in the International Journal of Knowledge-Based Development.