Anomaly detection using machine learning for ride-hailing companies

Ride-hailing services have established themselves as an essential part of travel. But recent reports suggest that the safety of the passenger and the driver can be at risk. Cab service companies have tried to address this concern by adding some features but these seem to be inadequate. By intergrating anomaly detection using machine learning into the surveillance systems, companies get to address many issues from monitoring driver/passenger behavior to proper accident management.

The number of crimes being committed in ride-hailing and taxicabs is on a steady rise. This is evident from the number of reports being filed by drivers for criminal assault and battery. From the drivers to the passengers, the safety concerns surrounding taxicab journeys are increasing by the decade. 

Recently, a drunken man attacked a Lyft driver in Florida, putting the cab driver in a chokehold from the backseat. In Delhi, six men broke a window of an Ola cab and assaulted the driver and passenger. The image below shows a few of the several such crimes recently reported in google news.

“It’s not a good feeling to have a gun to your head,” said Santanu Bose, a cab driver who lives in Rogers Park. He said he’s been robbed twice since becoming a licensed cab driver. “It did shake me up quite a lot for a few weeks.”

According to the Bureau of Labour statistics, workplace violence remains a leading source of occupational fatalities and injuries with taxicab drivers. Historically, they have experienced one of the highest homicide rates of any occupation. According to a study published by the American Journal of Preventive Medicine, the taxicab driver homicide rate was 7.4 per 100,000 drivers in 2010. 

The increasing number of fatalities is turning into a serious predicament for companies like Uber, Lyft, Ola and other ride hailing companies. This creates the need to enhance the security measures for better protection of cab drivers as well as passengers. Some of the major precautions that are being taken to ensure the safety of cab drivers today are as follows: 

• Barriers such as bullet-resistant glass between drivers and passengers prevent robberies, injuries and death.
• Security cameras record activities within the vehicle, discouraging violent behavior, and aiding in identifying passengers, if an assault does occur.
• Silent alarms (such as an external light) and/or radio communication allow drivers to safely request help.
• Vehicle tracking devices, such as global positioning satellite (GPS) systems, allow drivers in distress to be located.
• Improved lighting inside the taxi allows the driver to be aware of passenger behavior.

The most preferred safety precaution that organizations use is the installation of an in-cab surveillance camera. These cameras give perpetrators the feeling of being watched, which acts as a deterrent. It also increases the arrest rate of such criminals. 

Unfortunately, surveillance cameras are only used for investigative purposes. That is after the crime has been committed, law enforcement officials can playback the footage and analyze it to collect evidence. But, in today’s digitally empowered era, shouldn’t we expect more sophisticated solutions that can actually help prevent these crimes from happening in the first place?

This is where the relevance of anomaly detection using machine learing comes into play. 

Take a look at the below topics to know more:

• How anomaly detection using machine learning help us detect crimes?
• How an anomaly detection system works?
• Need to implement an anomaly detection system
• Benefits of an automated anomaly detection system
• Closing thoughts

How anomaly detection using machine learning help us detect crimes?

Anomaly detection is a technique that is used to identify items or events that do not conform to an expected pattern. Such techniques are used in situations where abnormal behavior needs to be identified.

By using machine learning-powered anomaly detection systems in taxi cabs, the occurrence of crimes can be detected in real-time. These systems can immediately identify the occurrence of anomalous activities such as assault, theft, rape, the presence of harmful objects or weapons, disputes, etc. Such a system can alert the concerned party, say, nearby police officials, to take necessary action. Timely alerts can help in more proactive engagement from the police department. 

 1. Real time alerts about mishaps

If any abnormal behavior is detected by the system, it immediately generates an alert or notification that can be sent to the nearest police department. This can help law enforcement authorities to take timely action, and diffuse the situation before it gets out of hand. 

2. Face identification to detect potential threats

The system can automatically identify the faces of passengers. It can use information from the company’s database of blacklisted passengers or public police databases of wanted criminals to identify any potential threats. If the system matches the face of a passenger to any of these databases, a notification will be sent to the concerned authority as well as notify the cab driver to stay safe.

Related article on How anomaly detection can help today’s business world

How an anomaly detection system works?

The video above shows the anomaly detection module of Emotyx – an AI-powered real-time video analytics suite. The real-time video feed from a surveillance camera is a series of video frames. For anomaly detection, the system performs classification models on every incoming frame, which will classify the frame into either an anomalous or non-anomalous class. Here, information captured from the previous frame is linked to the current active instance, to derive intelligent inferences.

Emotyx uses a proprietary hybrid neural network that has a core backbone net that generates high dimensional features from a set of continuous frames instead of every single frame. It also has a head net, which classifies the features generated by the backbone. 3D Convolutional Neural Networks (3D ConvNets) learn spatiotemporal features far better than the conventional 2D Convolutional Neural Networks (2D ConvNets). Orthodox 2D ConvNets run different filters on inputs both horizontally and vertically.

In layman’s terms, 2D ConvNets are capable of processing 2-dimensional inputs only. But the 3D ConvNets takes the time axis into its consideration along with the existing 2-dimensions. The 3D network which we designed can capture spatial information from the set of inputs. Emotyx batches a continuous bag of frames into one and passes it to the network to generate features. The output is sent to the head net, which is a linear classifier for the final classification task. The head net is shallow in nature but was able to fit well for the specific use case. 

To facilitate inter-communication among the two networks and for the final conversion of the classifier’s results, Emotyx uses different prime data structures like stacks and queues. The results of the classifier are stacked to a certain threshold to see whether there is a potential difference in the pattern or not. This difference might indicate a probable event of interest. Where in this case, detection of assault, theft, and other unfavorable events. The different metrics Emotyx uses to test the network include Area Under the Curve (AUC) and F1 score. 

AI-powered anomaly detection can be helpful in several other areas as well, such as public places, factory premises, prisons, etc. Here is a video of Emotyx’s anomaly detection module that can detect anomalies like an explosion, fights, robbery, etc.

Violence in taxicabs is still prevalent, despite the precautionary measures being taken by ride-hailing service companies like Uber and Ola. By using the AI-powered anomaly detection feature of Emotyx, it is possible to detect any violent behavior and send alerts to law enforcement authorities for corrective action.

Would you like to discuss more on this?

Need to implement an anomaly detection system

Take a look at the below points to gain better insight into the need for having an anomaly detection system to ensure the safety of passengers and drivers. 

• Monitoring mass area: The automated anomaly detection using machine learning can help you monitor vast areas that might pose a difficulty to monitor simultaneously. The system helps detect anomalies in crowded areas like malls where individual attention might not be possible to the human eye. It also helps to keep an eye on the deserted areas which are potentially unsafe. To ensure safety, monitoring the location and the activity of drivers and passengers at the time of the pick up can help. If any out-of-the-ordinary behavior is detected the authorities can be alerted against it for further action. 
• Monitoring detour behavior: Monitoring the route map of the journey and keeping a record of detour behavior to alert passengers and authorities can ensure safety. This can also help track the people they have contact with on-road for further investigation. 
• Accident monitoring: In the event of an accident, the last recorded footage in an anomaly detection system can assist the authorities and the company in understanding the cause of the accident. This can help them take actions accordingly and also detect the location of the reported accident of that assistance can be sent immediately to the accurate location. Additionally, it can help detect fraudulent or false claims in an unfortunate event. The footage can be proof that helps the real victim. 
• Occupancy monitoring: Automated anomaly detection using machine learning can monitor the occupancy in the vehicle and monitor the activities and behaviors of the occupants. In recent times this can be used to monitor the covid regulations defaulters. It can also be used to detect other activities that can show struggle or misbehavior. 
• Monitoring rash driving: Anomaly detection using machine learning can also detect rash and negligent driving on the roads as well as disregarding other regulations. 
• Create an award system: With the data of monitored driving performance and behavior towards the customers, drivers can be rewarded to encourage this safety-oriented and punctual nature. 

Benefits of an automated anomaly detection system

• In real-time identification of anomalies can help point out all issues that deviate from normal behavior autonomously. 
• The AI can detect patterns within the various groups of different anomalies using correlation thus making it easier to find the root cause. 
• It reduces the effort and time to detect and understand anomalies hence pushing toward efficient and better decision making. 
• With the use of unsupervised algorithms, understanding things like periodicity, and seasonality in the data can help find further patterns. 
• By detecting maximum occupancy within the vehicle and the behavior of the passengers, it can help the company detect any anomalies in the pattern of normality. 
• Since the system can allow different data sources footages of accidents or other anomalies in the pattern can be sent over to law enforcement authorities.

Related article on How machine learning is improving our fraud detection systems

Closing thoughts 

Automated anomaly detection using machine learning helps to eliminate the concerns around security and safety in ride-hailing services. The systems provide better insights into the root causes and thus help the authorities take preventive measures to avoid the detected anomalies. Ensuring the safety of the passengers and the driver can lead to a better experience for both employees and the customers of the ride-hailing companies thus helping to boost overall productivity.