- Our inspiration for this challenge was to help people in very short duration of time with having 99% of accuracy & success rate with our predictive modelling
What it does
- Our program takes the values logged by CCMS, cleans it, scans the reformatted data for irregularities (kwh), compares it with standard values of street lights logged by the CCMS and the classifies it further into faulty or working classes. After classification our prediction model then takes the values logged by CCMS and predicts the likelihood of a streetlight failing with its accuracy, Currently our trained model is > 97.19% accurate, having a validation accuracy of > 99%.
How we built it
- We created our own algorithm for comparing and classifying the values as faulty or working streetlight classification. And we used 'Keras' the open-source Python Deep Learning library for creating our convolutional neural network for predictive modelling.
Challenges we ran into
- We were at first having problems with inconsistencies in the data set but after refactoring and cleaning it we were able to perform our analysis on it and were then able to create our machine learning model, predictive analysis.
Accomplishments that we're proud of
- We were able to create our machine learning model having an accuracy of more than 97% with validation accuracy of over 99%.
- We were able to correctly classify specific excerpts from the logged data as faulty, which was our challenge for this theme.
What we learned
- We learned to clean, refactor data set, create a machine learning model on it and use it for a practical application.
What's next for Predictive Analysts
- Our next goal is to uniquely identify streetlights and their location separately. Thus, reducing the whole maintenance costs, better detection of faulty streetlights and reducing the time taken to resolve the issue as well.