Description of the process of using construction equipment productivity data, weather data and machine learning algorithms to quantify and predict the effect of weather on productivity levels. No data or code included as it is company IP
Equipment data present is sent by IoT gateway/telematics devices that collect data from various sensors installed on the equpiment.
Imported equipment data (usage, location, etc) from an SQL server Data Warehouse and weather data from the DarkSky API, using latlongs as parameters.
Data was combined by importing equipment data from the SQL server warehourse into Microsoft PowerBI and creating an additional custom column that stores the response of a DarkSky API call with location from other columns in the table as parameters.
-> Calculated productivity-per-day from various usage parameters
-> Isolated only productivity and other weather columns
-> Used DarkSky's "icon" datapoint which summarizes the weather that day into categories like "clear", "partly cloudy", "cloudy", etc
-> Used DarkSky's "apparentAverageTemp" datapoint to obtain realistic temperature levels
-> Split the dataset into 80% training and 20% test
-> Ran various ML algorithms on the training set with only weather as features to evaluate the best, settled on Random Forest
-> Produced low levels of accuracy which were unacceptable
-> Realised that the stage that a construction project is currently in has an impact on equipment utilization (some stages of the project are more equipment dependent than others)
For example, equipment are highly utilized during the beginning and middle stages of the project, less during the end of a project
-> Added a feature "projectProgressPercentage" to denote the stage of the project
-> Cross-validation testing on the training set now produced a reasonably acceptable accuracy of ~75%, given the fact that I didn't have much data
-> Similar accuracy levels were observed on the test set
Ran the prediction again on the entire dataset to predict productivity levels for various temperatures and weather conditions (x axis - temperature/conditions and y axis - productivity per day)
