[Pycon] [new paper] "Luca Bravi" - Automatic identification of anomalous shutdown in Vehicle Tracking Units: a practical machine learn

[info a pycon.it]

Title: Automatic identification of anomalous shutdown in Vehicle Tracking Units: a practical machine learn
Duration: 45 (includes Q&A)
Q&A Session: 15
Language: en
Type: Talk

Abstract: Vehicle Tracking Units (VTUs) are becoming increasingly popular due to the pervasiveness of their applications: navigation, fleet management systems, insurance,
road-side assistance are some examples. For these purposes it is essential that VTUs are continuously and properly working and thus malfunctioning VTUs must be promptly recognized and fixed. A complete stop in reporting is the most common malfunction and can have several reasons: bad wiring, GSM connection problems, broken unit, etc. However a VTU could stop to report also for non anomalous reasons such as battery charge saving, underground parking or poor cellular coverage. Being able to automatically discriminate between these two cases is crucial to decide whether a technical intervention is needed.
In this talk we show how we tackled this real world problem building a Random Forest Machine Learning model based on Verizon Connect data, describing all the steps of the pipeline from data cleaning and feature engineering to performance evaluation using the sci-kit learn and pandas python libraries.

Tags: [u'scikit-learn', u'Data Mining', u'Machine Learning', u'bigdata', u'analytics', u'pandas']