Manhattan condominiums with similar sizes, and designs are compared against one another. Each unit’s location, market value, and attributes are analyzed to estimate its potential rental income.
- Source of data:
- Brief description of data
NYC Condominiums with similar features are compared and assessed as rental units.
- What is the target?
Manhattan Condominimum Est. Gross Income
- Is this a classification or regression problem?
This is regression problem as the target value is not limited to a finite number of classes or values.
- How many features?
46-1(target) columns = 45 features (TBD). I will undoubtedly drop several more columns once I begin to look through the data more closely.
- How many rows of data.
1068
- What, if any, challenges do you foresee in cleaning, exploring, or modeling with this dataset?
This dataset has a lot of columns, the challenge lies in determining which columns will be the most effective for predicting our target values.
DOF Condominium Rental Analysis Data Dictionary.xlsx
A color coded map makes distinctions between areas clear and easy to interpret. The highest market values (represented by warming purple to yellow hues) are concentrated towards the center of the island between mid thirty streets and the mid sixties. Prices appear to decrease linearly as we move further north, away from Central Park's northern bound; Central Park North.
The graph above provides a summary of the most and least expensive neighborghoods on the island of Manhattan. We can see that the Midown Central Business District (CBD) is the most expensive and Inwood at the northern most point of the island is the least expensive.
Machine learning is used to streamline, expedite and enhance the forecasting process by removing repetitive manual procedures. Automating algorithms or programs allow for higher efficiency, deeper insights, and higher profitability. In the Machine Learning section, we use a copy of the unadulterated data as a basis and instructive training tool for our models. Several different models will used with this training data to draw inferences and detail elemental relationships between features or columns so that they will be able to interact with, and perform similar analysis on testing or unseen data.
Our chosen model, at its best, is producing errors that are close to 200 thousand dollars. We recommend employing it as an aid or guide in selecting several properties for viewing, but not as the sole fulcrum for purchasing decisions.
We suggest that our client focus their time and resources on Condominiums located in the center of the island or ones that are in close proximity to the Hudson or East River. Units on the south and west of the island seem have higher market values and therefore higher potential rental income.
The inclusion of more physical features would drastically increase the predictive power of our model. Physical characteristics like bedroom and bath counts, ceiling and window heights, exterior square foot, heating and mechanical systems, Wi-Fi and app friendly controls, exposure direction and linear feet would be great additions to our data set. National and Global Considerations like CPI, Fed fund rate, market indexes, home supply, and international shipping metrics, would also be insightful.