We have some exciting news to share with you about the FleetTrackr application. We know that receiving notifications or alerts can sometimes be overwhelming and distract us from our daily routines, especially if they are irrelevant to our needs or serve no real purpose.

With this in mind, we have incorporated new machine learning (ML) models that are designed to generate meaningful alerts, filter alerts, and learn from user behavior. These ML models can help to significantly reduce the number of unnecessary notifications, enabling you to focus on what is important.

Let's dive in and learn a little more about these models.

Edge ML models

Idle-Time-Detector

‍This model monitors a device's usage patterns and determines when it is in an idle or near-idle state based on CPU, RAM, and power consumption. This information is used to determine when it is best to update the device. For example, the model may learn that Tuesdays through Fridays at 10:00 PM have the least amount of activity on the device, making it a good time to send alerts to perform maintenance tasks. The Idle-Time-Detector model can notify users about when their device is typically the least active, enabling them to make informed decisions about when to schedule updates or maintenance.

‍

TegraStats-Anomaly-Detector

‍This model monitors a device's CPU and RAM usage, temperature, and disk usage. During the pilot phase, the system learns each device's usage patterns and generates a unique model for it, as different devices may run different software.This enables it to determine what is normal for the device and alert the user when an anomaly occurs, such as a crash or a sudden change in usage. This model can alert users about any unusual behavior that could lead to future crashes or malfunctions.

‍

The Physical-Disturbance-Detector

‍This model makes use of the IMU module found in all devices in detecting physical disturbances to devices and alerting users in real time. The IMU module includes a gyroscope, accelerometer, and magnetometer for detecting changes in acceleration, orientation, and magnetic fields. These sensors work together to detect physical movement or changes in the environment of the device. For example, if a camera falls off its mounting point or a device is moved from a shelf, the model detects the changes and sends an alert to FleetTrackr.

‍

‍

Server ML model

Alert Priority Classifier

‍This model on the FleetTrackr server provides users with a more personalized experience by allowing users to select and prioritize the alerts that are most important to them. The FleetTrackr server receives many alerts from devices, which can overload it. To address this, the Alert Priority Classifier has a pilot phase where users can select their preferred alerts. The model then learns the user's preferences and prioritizes alerts that the user wants to see more often. For example, if a user is uninterested in disc usage alerts, the system reduces their frequency and priority. If the user is interested in CPU usage alerts, they are given a higher priority and displayed more frequently.

‍

‍

These are just a few of the many ML models that we employ to make FleetTrackr even more user-friendly and efficient. At SmartCow, we are committed to making technology work for you and not the other way around. Thank you for your continued support and for choosing FleetTrackr.