Yun.Bun

---

Define, Build, Deliver

Here’s a neat fact – adopted dogs tend to suffer less from separation anxiety than those bought from breeds or pet stores!

Below is a pipeline for addressing a real-world problem using data science techniques and relevant libraries, demonstrated through a dog adoption use case.

By applying these steps with the appropriate tools and programming libraries, shelters can streamline their operations and enhance the efficiency & timing of the dog adoption process!

---

1st Problem Definition:

• Goal: Predict the likelihood of a dog being adopted from an animal shelter based on its characteristics.
• Objective: Build a model that helps shelters identify which dogs have a higher chance of adoption, so they can focus more efforts on those who are less likely to be adopted.

2nd Data Collection & Acquisition:

Source: Gather data from shelters or adoption platforms (e.g., Petfinder, local shelters).

Features:

• Breed (e.g., Labrador, Beagle)
• Age (e.g., 1 year, 3 years)
• Size (e.g., small, medium, large)
• Temperament (e.g., friendly, anxious, energetic)
• Health (e.g., vaccinated, special needs)
• Shelter duration (how long the dog has been in the shelter)
• Adoption history (whether the dog has been adopted before)

Data Sources: Shelter databases, adoption websites, or API integrations from platforms like Petfinder.

How might this look like?

PyTutor

dog1 = {
"Breed": "Labrador",
"Age": 2,
"Size": "Large",
"Temperament": "Friendly",
"Health": "Vaccinated",
"Shelter Duration (days)": 30,
"Adoption History": False
} Want access? Visit Yun.Bun I/O

3rd Data Cleaning & Preprocessing:

Missing Values: Some dogs might have missing values in features like age or breed. We would either fill them with appropriate defaults (e.g., average age) or drop those rows.

Outliers: Identify any data points that don’t make sense, such as extremely young dogs listed as 100 years old.

Categorical Variables: Convert categorical variables like breed or temperament into numerical representations (using one-hot encoding or label encoding).

Normalization/Scaling: Scale numerical values like age or weight so that larger values don’t dominate the model.

4th Exploratory Data Analysis (EDA):

Visualizations:

• Distribution of Dog Ages: How old are the most commonly adopted dogs?
• Breed vs. Adoption Rate: Which breeds tend to get adopted faster?
• Correlation: Are there any patterns between dog size and adoption likelihood? Do dogs with special needs have lower chances of adoption?

Insights: Through EDA, we might discover that medium-sized dogs are adopted faster, or that dogs with “friendly” temperaments have a higher likelihood of adoption.

5th Feature Engineering & Selection:

Feature Creation: Combine multiple features into a single one (e.g., combining breed and size to create a “breed_size” category). Create binary features, such as “has_vaccination” or “special_needs”.

Feature Selection: Use techniques like correlation analysis or feature importance (e.g., from decision trees) to select the most relevant features for the model. Perhaps “age” and “temperament” are more important than “health” in predicting adoption likelihood.

PyTutor

# Step 1: Original shelter dog data
dog1 = {"Breed": "Labrador", "Age": 2, "Size": "Large", "Temperament": "Friendly", "Health": "Vaccinated"}
dog2 = {"Breed": "Beagle", "Age": 3, "Size": "Medium", "Temperament": "Energetic", "Health": "Special Needs"}
dog3 = {"Breed": "Poodle", "Age": 1, "Size": "Small", "Temperament": "Anxious", "Health": "Vaccinated"}

dogs = [dog1, dog2, dog3]
# Want access? Visit Yun.Bun I/O

6th Modeling & Evaluation:

• Model Selection: Try different machine learning models such as: Logistic Regression: Predicting the probability of adoption. Random Forest: Understanding feature importance and making predictions. XGBoost: A more advanced, accurate model.

• Model Evaluation: Accuracy: How well the model predicts the adoption status. Precision and Recall: We care about minimizing false positives (dogs predicted as adoptable but aren’t) and false negatives (dogs predicted as not adoptable but are).ROC Curve: Evaluate the model’s ability to distinguish between adopted vs. not adopted dogs.

7th Model Deployment & Monitoring

How might this look like?

Deployment: Deploy the model as a service that shelter staff can use to predict adoption likelihood.

The model could be integrated into an app or website where shelters input dog information and get adoption likelihood predictions.

Monitoring: Continuously monitor the model’s performance over time. For instance, if the model starts to predict inaccurately due to new dog breeds or changes in adoption patterns, we may need to retrain the model.

8th Visualization & Reporting:

Dashboards: Create a user-friendly dashboard that allows shelter staff to see the model’s predictions for each dog and track adoption progress over time.

Reports: Generate periodic reports showing which dogs are likely to be adopted soon and which need more attention (e.g., more photos, special events, or foster programs).

Interactive Visuals: Use charts to show correlations between adoption likelihood and various features (e.g., adoption by breed, age, or temperament).

---

Why is data handy in this context?

• Data helps shelters track outcomes, like how long dogs stay before adoption, their return rates, and which profiles are most appealing to potential adopters.
• Therefore, approaches can be tailored to boost adoption success and cut down on overcrowding!

Thanks for reading!