My inspiration was to challenge myself and push my limits as a data science master's student. Participating in a data science challenge provides me with an opportunity to apply my skills and knowledge in a practical setting. It allows me to showcase my problem-solving abilities and analytical skills while working with real-world datasets. I am motivated to learn from other participants, explore different approaches, and gain valuable insights that will contribute to my growth as a data scientist.
FaultForecast utilizes historical data, network monitoring, and advanced algorithms to analyze patterns and identify early indicators of potential service faults. By predicting these faults in advance, Frontier Communications can take proactive measures to address issues, minimize service disruptions, and enhance customer satisfaction.
FaultForecast was built using a combination of data engineering, machine learning, and software development techniques. We collected and processed large volumes of historical data, performed feature engineering to extract relevant information, and trained predictive models using advanced algorithms. The Gradient boosted model is used here which is deployed using the streamlit and github.
During the development of FaultForecast, we encountered challenges such as handling large and complex datasets, ensuring data quality and consistency, and fine-tuning the predictive models for optimal performance. The Major challenge was identifying the relavant features that contribute to the service shortage.
We are proud to have developed FaultForecast, a powerful tool that can help Frontier Communications proactively address service repairs. Our team was successfully above to make a model that performs with an accuracy of 97 percentage without overfitting on the data.
Throughout the development of FaultForecast, we gained valuable insights into the importance of data preprocessing, feature engineering, and model selection. Several model were considered for the task. I was able to make an end to end from data preparation to model deployment.
Moving forward, we plan to further enhance FaultForecast by incorporating real-time data streams and integrating it with Frontier Communications' existing monitoring systems. We aim to include the time series aspect to the model as well so that when there is consequent requires from customers we will be able to identify the network storage with a better understannding.