maejok / clustering_for_single_family_residence_investors_in-ga Goto Github PK
View Code? Open in Web Editor NEW### About this Project This is a partnership research between NYC Data Science Academy and HaystackAI. We used the latest Machine Learning sk-learn models for descriptive data analysis in Python. Objective of The Project: To provide insight on how to use clustering to identify outliers in the housing market, characterise and identify different kinds of anomalies, and identify opportunities for investment in each neighbourhood. To provide data backed insight for Single family Rensidence( SFR) PropTech investor who wish to invest in a place with long term growth by highlighting different areas with increased gentrification. To Provide diverse metrics to identify unappreciated opportunities and enable real estate professionals to be ahead of the market . The Motivation of The Project: The traditional use of ZipCodes for demarcation of local areas have limited use in the business context as there can be different zip codes that represent the same market and vice versa. We try to use Machine learning Clustering techniques to determine the natural group of clusters for Single Family Residence. The Data: The data used for this project was provided by HaystackAI and protected under extant copy right laws. Other alternative sources of data that are publicly available were from Broker Listing Data,Crime_Diary, Census, Local News, Amenities, Finance and were downloaded from the assessors website as text files and contained both categorical and continuous data. We used unsupervised learning method for this project, created clusters of properties with geocoded locations to identify properties and their locations with similarities over a large number of features. Principal Component Analysis PCA, was used for dimensionality reduction, and unsupervised clustering was performed using k-means, hierarchical agglomerative clustering (HCA) and Density-Based Spatial Clustering of Applications with Noise (DBSCAN) using scikit-learn libraries. We identified seven natural clusters with unique descriptions and diverse investment potentials. We tested different clustering techniques including Kmeans with different initialisations, MiniBatchKmeans, Agglomerative Hierarchical Clustering and DBSCAN. We used the Elbow Method and Silhouette scores metrics for determining the optimum number of clusters. Cluster 1 , for example, has these characteristics, lower cost, very high crime rate, very high rate of growth, fewer amenities. It is likely attractive for investors that target lower income buyers and who would like to maximise ROI. Similarly, Cluster 3 has these characteristics, lower cost, stable crime rate, very high rate of growth-good amenities. It is ideal for medium term investors.
License: MIT License