Machine Learning - SF Fire Department False Alarms

Project Overview

• Large dataset holding ro 4 million records, for emergency calls from years 2000-2016
• Prepare data and assemble features for ML models
• Hypertune paramters to compare and select the most accurate model
• Pyspark is the programming language used
• You can access the Jupiter Notebook at this LINK

Objectives

Design an accurate machine learning model that can be used to predict future false alarms. This will help increase the efficiency of the fire department, detecting false alarms prior to expending resources.

Data Preperation

Data is accessed through a public repository on aws s3, uploaded by the government of San Fransisco.

The s3 drive is mounted and tables ‘Fire_Department_Call’ and ‘Fire_Incidents’ are extracted. From here, both tables are joined by ‘IncidentNumber’ creating a single dataframe object. Vector features are then selected and assembled for the machine learning pipeline.

Data Modeling

4 Different Machine Learning models are constructed and evaluated

• Decision Tree
• Random Forest
• Linear Regression
• Logistic Regression

You can access the Jupiter Notebook at this LINK

Conclusions

• The Elastic Net model is the best, and will be selected for future anaylsis.