Silicon Valley Data Science: Improving Train Arrival Predictions using TensorFlow

Silicon Valley Data Science recently developed an image-recognition sensor that can improve the accuracy of Caltrain arrival predictions. This was achieved using only $130 worth of hardware, with the sensor’s code relying on open-source software libraries.

SVDS had experimented with other approaches to this problem before, but with limited success. The previous attempts at improving the accuracy of train arrival times failed due to an inability to differentiate between one type of train and another. However, with the addition of TensorFlow, a machine-learning software library, this goal proved much more feasible.

TensorFlow is an open-source software library that provides users with documentation resources, and is frequently used in machine learning research and engineering. Google uses TensorFlow technology in many of their speech recognition applications, as well as with the Google Brain project.

The addition of TensorFlow allowed the new SVDS trainspotting sensors to differentiate between different types of objects, more specifically, the subtle differences between Caltrain trains, and non-Caltrain trains. The sensors correctly differentiated between moving Caltrains and other trains 89.35 percent of the time after training with the TensorFlow software libraries – a significant improvement from where the sensors started.

Innovative developments like this will likely grow more common as software libraries begin to expand, and as hardware becomes more readily available. The incremental changes taking place with the help of global software libraries will help encourage development in nearly all fields.

Silicon Valley Data Science: Improving Train Arrival Predictions using TensorFlow
Trend Reports

10 Trends to Watch

What are the technology and user trends that will shape your business?