Continuing on the loan allocation problem, it is eminent that alternative machine learning methods be employed that do not necessarily require supervision to predict the likelihood of a customer paying or defaulting on their loan.
Herein, we shall employ several clustering technique to previously used data to try and ascertain this likelihood.
Our modelling approach intends to create a clustering model that can perform well with at most four categorical clusters each for a credit score between 0 and 1.
For this projects, based models were developed and fitted on a contrived data set. These were; affinity propagation, agglomerative clustering, BIRCH, DBSCAN, gaussian mixture, k_means, mean_shift, mini batch k means, OPTICS and spectral clustering.
The intend, if to find which of these models will best fit our data and give definitive clusters for our statement problem.
The first model - performed averagely, but was unable to give an anticipated definitive outcome. Scored an inertia value of 4406 with a minimum of 7 clusters.
Predicted two distinct cluster but imbalanced clusters both for the original dataset and the synthetically generated data. The synthetic data created a more distinct clustering.
Predicted two less distinct clusters for the original data. Predicted clusters were more distinct on the synthetic data.
Predicted two less distinct clusters for the original dataset. Predicted more distict clusters on the synthetic data.
Other clusters deployed in this project that did not perform well on creating or identifying proper clusters: OPTICS, K-means, mini-batch-k-means, mean-shift and spectral clustering.