Capture trends and insights by visualizing the variables and distribution of service providers, technology, tiers and KPI for broadband services in different states of Australia.
Deep dive into the KPI by acquiring the probabilities of classification values that sets the standards of being impaired and underperforming
Applied accuracy, precision, recall, f1, jaccard and log loss scores to determine how good our model fits our data set.
Implemented dimensionality reduction technique to summarize the large data set increasing interpretability while keeping the significant information.
Utilized K-Elbow method and visualization to determine the optimal number of clusters that suits the data sample.
Creates the hierarchical segmentation that make use of the variables to set profiles for each clusters.
Analysis of retail service providers versus other categorical variables such as technology, tier, and state.
RSP vs. Technology
RSP vs. Tier
RSP vs. State
KPI investigation in terms of download speed, upload speed and latency aggregated values.
RSP vs. Maximum KPIs
RSP vs. Minimum KPIs
Technology vs. Mean KPI
Classification of connection's impairedness and underperformance based on KPI metrics for 100/20 Mbps. Extracting the probabilities of each KPI values to determine the standard value.
With 89.5% accuracy score, the limits of the metrics to consider our service is impaired are:
With 92.1% accuracy score, the limits of the metrics to consider our service underperforming are:
Cluster profiles developed using agglomerative clustering and k-elbow method.
Cluster 0:
Cluster 1:
Cluster 2:
Cluster 3: