Answer:
To predict homelessness within the next 90 days using the given data, a predictive model could apply a logistic regression analysis, decision trees, or machine learning techniques such as random forests or gradient boosting. The model should include key variables such as age, sex, educational attainment, and substance abuse history (e.g., heroin, fentanyl, meth, cocaine) to assess risk factors associated with homelessness. Data preprocessing steps like normalization, handling missing values, and feature selection are crucial for model accuracy. Training the model on historical data with labeled outcomes (homeless or not) and then evaluating its performance using metrics like accuracy, precision, and recall will help refine its predictive capabilities. Continuous monitoring and adapting the model with new data will improve predictions over time, ensuring it stays relevant and accurate.
gemini / gemini-2.5-flash
Score: 7.5
The answer correctly identifies appropriate machine learning models (logistic regression, decision trees, random forests, gradient boosting) and essential steps like data preprocessing, training, and evaluation metrics. However, it fails to explicitly include two highly relevant variables provided in the prompt: "crediscore" and "zip code," which are crucial indicators of financial stability and neighborhood risk. A significant omission is the lack of any discussion regarding the ethical implications, potential for bias, and responsible deployment of a model predicting homelessness, especially when using sensitive data like substance abuse history. This is a critical consideration for an AI agent operating in such a domain.
anthropic / claude-sonnet-4-5-20250929
Score: 6.5
The answer is technically sound but incomplete and misses critical considerations. It correctly identifies appropriate ML algorithms (logistic regression, random forests, gradient boosting) and mentions important preprocessing steps. However, it has significant gaps: (1) Doesn't address the severe ethical concerns and potential discrimination in using sensitive attributes like addiction history, race-correlated zip codes, and health data for homelessness prediction; (2) Fails to mention the need for informed consent and data privacy protections; (3) Doesn't discuss class imbalance issues (homelessness is relatively rare); (4) Omits important predictive features like income, employment status, housing stability, eviction history, and social support networks; (5) Doesn't address how the model would actually be used ethically or what interventions would follow predictions. The technical approach is reasonable but the answer lacks crucial context about responsible AI deployment in social services.
cohere / command-r-plus-08-2024
Score: 8.5
The answer is comprehensive and accurate, providing a solid strategy for predicting homelessness within 90 days. It suggests relevant machine learning techniques, essential data preprocessing steps, and performance evaluation metrics. However, it could be improved by specifying the importance of feature engineering, especially for categorical variables like sex and educational attainment. Encoding these variables appropriately is crucial for model accuracy. Additionally, the answer could emphasize the potential challenges of imbalanced data, as homelessness might be a relatively rare event in the dataset, affecting the model's performance. Techniques like oversampling, undersampling, or using class weights could be mentioned to address this issue. Lastly, while continuous monitoring is suggested, the answer could highlight the need for regular model validation and recalibration to maintain its predictive power over time.
Scores are 0–10. The selected AI’s score is a self-rating.
