Social Science Research/Rehabilitation of the Homeless

Social Science Research Rehabilitation of the Homeless Quality of Life and Mental Health in Disabled People Illnesses, Addictions, and Care for the Homeless The study https://arxiv.org/abs/2302.04455 titled "Rehabilitating Homeless: Dataset and Key Insights," introduces a new and extensive dataset on homeless rehabilitation, collected over two decades by a Russian nonprofit organization. It is described as the first dataset to contain detailed information on thousands of homeless individuals seeking rehabilitation. The paper presents the results obtained from applying machine learning models to predict rehabilitation success, highlighting the importance of certain characteristics in predicting outcomes. The primary objective of the data analysis is to predict the value of the variable contract_success related to the completion of the "contract," which represents the agreement between the client and the organization on rehabilitation goals, using the other variables in the dataset. A successfully completed contract is defined as "fulfilled." The dataset also includes binary variables (0 or 1) for 43 contract types, representing the different specific goals the homeless person intended to achieve through the contract with the organization. The data and the authors' Python analysis can be downloaded here: https://github.com/LEYADEV/homeless Dataset Description The dataset variables with their descriptions are as follows: Variable Description type id Contract type ID (43 types total) gender client's gender age client's age in years is homeless the client is homeless is married client's marital status has dependants the client has dependents has profession the client has a profession has residence The client has residency (but may be unable to use it for any reason) contacts relatives The client has recently contacted relatives source perm job The client's primary source of income is a permanent job source temp job The client's primary source of income is a temporary job source pension The client's primary source of income is a pension source alms The client's primary source of income is alms source relatives The client's primary source of income is relatives source other The client's primary source of income is not listed source gathering The client's primary source of income is collection source charity The client's primary source of income is charity source rehabs The client's primary source of income is rehab centers is citizen the client is a citizen of the country is retired the client is retired disability group the client has a disability in a certain group (1, 2, or 3) is registered the client is registered from orphanage the client lived in an orphanage was confined the client was in isolation chem confirmed the client confirms a chemical dependency chem possible possible chemical dependency chem no answer the client refused to answer about the chemical dependency chem no the client denies chemical dependency contract_success 0 - if the contract is "partially fulfilled due to the client's refusal" or "not fulfilled due to the client's refusal" or "partially fulfilled due to the client's absence" or "not fulfilled due to the client's absence"; 1 - if the contract is "fulfilled" comments alcohol The client has a problem with alcohol comments chem The client has problems with chemicals or medications comments disability The client has a disability comments absence The client does not see a social worker comments orphange The client grew up in an orphanage comments work The client is looking for work comments confinement The client was in solitary confinement comments noresidence The client has no place to live comments aggression The client displays aggressive behavior comments disturbance The client has a disturbance in a shelter comments disease The client has a disease comments death The client, or someone close to the client, has died comments nursinghome The client needs to be admitted to a nursing home comments relatives The client has relatives comments illiteracy The client is illiterate comments foreigner The client is a foreigner comments ticket The client needs a ticket home comments court The client is having trouble with a court comments pregnancy The client is/was pregnant comments mental The client has a mental problem comments fire TheClient's residence has been burned down comments social benefits Client needs a pension or other social benefits comments divorce divorced client type 1 Restore/obtain passport type 2 Apply for social benefits type 3 Obtain medical insurance type 4 Restore/obtain tax identification number (certificate) type 5 Obtain documentation to support work experience type 6 Register disability type 7 Arrange for nursing home placement type 8 Temporary registration with organization type 9 Registration at another address type 10 Employment type 11 Employment with accommodation type 12 Arrange for a night shelter type 13 Alcohol and drug rehabilitation centers type 14 Arrange for a hospital type 15 Search for relatives type 16 Return to hometown type 17 Determine/assume citizenship type 18 Legal assistance in court type 19 Obtain housing type 20 Cancel fraudulent transaction and restore housing type 21 Submit request(s) and/or application(s) type 22 Rehabilitation program (name omitted) to preserve anonymity type 23 Restore school records type 24 Issuance of a duplicate birth certificate or other document type 25 Obtain/reinstate military ID type 26 Obtain an international passport type 27 Rehabilitation equipment (prosthetic devices, etc.) type 28 Obtain social insurance certificate (insurance - individual personal account number) type 29 Arrange for placement in a social adjustment center type 30 Assistance with vocational training type 31 Identification type 32 Registration at an AIDS center type 33 Arrange for placement in a crisis center type 34 Psychological support type 35 Recalculate/reinstate/obtain social benefits type 36 Complete case in civil court type 37 Obtain unpaid wages upon dismissal type 38 Health Restoration Type 39 Employment Adjustment Program Type 40 Supported Housing Type 41 Support for the Rehabilitation Program for Clients with Chemical Dependency Type 42 Pre-Housing Preparation Type 43 Hostel Impact of the dataset variables on contract success The impact of alcohol consumption on contract success is negative. In fact, by calculating the inference of the two variables, we reject the null hypothesis H0 and can state that we are 95% confident that the percentage of customers who consume alcohol and fulfill the contract is lower than the percentage of customers who do not consume alcohol and fulfill the contract, given that p_value < 0.05: library(statsr) #Loading data: df <- read.csv("final_with_onehot_types_ALL_221120.csv") df$contract_success <- factor(df$contract_success, levels = c(0,1), labels = c("not fulfilled","fulfilled")) df$comments_alcohol <- factor(df$comments_alcohol, levels = c(0,1), labels = c("No","Yes")) df <- df[,-c(49,53)] inference(y = contract_success , x = comments_alcohol, data = df, statistic = "proportion", type = "ht", null=0,alternative = "greater", method = "theoretical", success = "fulfilled" ) Response variable: categorical (2 levels, success: fulfilled) Explanatory variable: categorical (2 levels) n_No = 6351, p_hat_No = 0.498 n_Yes = 1841, p_hat_Yes = 0.447 H0: p_No = p_Yes HA: p_No > p_Yes z = 3.8543 p_value = 1e-04 Similarly, by calculating inferences for all other variables with the variable contract_success, we can evaluate whether the impact on the success of the contract is positive or negative: for (i in 13:98) { dat<-inference(y = contract_success, x = as.factor(df[,i]), data = df, statistic = "proportion", type = "ht", null=0, alternative = "greater", method = "theoretical", success = "fulfilled", verbose = FALSE ) if (dat$p_value<0.05) print(paste(colnames(df[i]),"negative impact")) else print(paste(colnames(df[i]),"positive impact")) } [1] "comments_alcohol negative impact" [1] "comments_chem positive impact" [1] "comments_disability positive impact" [1] "comments_absence negative impact" [1] "comments_orphange negative impact" [1] "comments_work positive impact" [1] "comments_confinement negative impact" [1] "comments_noresidence positive impact" [1] "comments_aggression positive impact" [1] "comments_disturbance negative impact" [1] "comments_disease positive impact" [1] "comments_death negative impact" [1] "comments_nursinghome positive impact" [1] "comments_relatives negative impact" [1] "comments_illiteracy impact" to positive" [1] "comments_foreigner negative impact" [1] "comments_ticket positive impact" [1] "comments_court positive impact" [1] "comments_pregnancy positive impact" [1] "comments_mental positive impact" [1] "comments_fire negative impact" [1] "comments_social_benefits positive impact" [1] "comments_divorce positive impact" [1] "contacts_relatives positive impact" [1] "no_relatives positive impact" [1] "source_perm_job positive impact" [1] "source_temp_job negative impact" [1] "source_social_benefits negative impact" [1] "source_alms negative impact" [1] "source_relatives positive impact" [1] "source_other positive impact" [1] "source_gathering positive impact" [1] "source_charity positive impact" [1] "source_rehabs negative impact" [1] "is_citizen positive impact" [1] "is_retired positive impact" [1] "is_registered positive impact" [1] "from_orphanage positive impact" [1] "was_confined positive impact" [1] "chem_confirmed positive impact" [1] "chem_possible positive impact" [1] "chem_noanswer positive impact" [1] "chem_no positive impact" [1] "type_1 positive impact" [1] "type_2 negative impact" [1] "type_3 negative impact" [1] "type_4 positive impact" [1] "type_5 positive impact" [1] "type_6 negative impact" [1] "type_7 negative impact" [1] "type_8 positive impact" [1] "type_9 positive impact" [1] "type_10 positive impact" [1] "type_11 positive impact" [1] "type_12 positive impact" [1] "type_13 positive impact" [1] "type_14 positive impact" [1] "type_15 positive impact" [1] "type_16 positive impact" [1] "type_17 negative impact" [1] "type_18 negative impact" [1] "type_19 negative impact" [1] "type_20 negative impact" [1] "type_21 negative impact" [1] "type_22 positive impact" [1] "type_23 positive impact" [1] "type_24 positive impact" [1] "type_25 positive impact" [1] "type_26 positive impact" [1] "type_27 positive impact" [1] "type_28 positive impact" [1] "type_29 positive impact" [1] "type_30 positive impact" [1] "type_31 negative impact" [1] "type_32 positive impact" [1] "type_33 positive impact" [1] "type_34 negative impact" [1] "type_35 positive impact" [1] "type_36 negative impact" [1] "type_37 positive impact" [1] "type_38 negative impact" [1] "type_39 negative impact" [1] "type_40 positive impact" [1] "type_41 positive impact" [1] "type_42 positive impact" [1] "type_43 positive impact" Predicting Contract Success To predict the contract_success variable using the others, it is best to implement the XGBoost algorithm, obtaining a forecast accuracy of 77% (better results can be found in the study authors' Jupyter report): import pandas as pd from sklearn.model_selection import train_test_split from sklearn.metrics import classification_report from xgboost import XGBClassifier with_onehot_types = pd.read_csv('final_with_onehot_types_ALL_221120.csv') x_columns_t = list(set(with_onehot_types.columns) - set(['contract_id', 'client_id', 'contract_success', 'contract_status_id'])) X_t = with_onehot_types[x_columns_t] y_t = with_onehot_types.contract_success X_train_t, X_val_t, y_train_t, y_val_t = train_test_split(X_t, y_t, test_size=0.2, stratify=y_t) model = XGBClassifier() model.fit(X_train_t, y_train_t) preds= model.predict(X_val_t) print(classification_report(y_val_t, preds)) precision recall f1-score support 0 0.77 0.78 0.78 883 1 0.76 0.75 0.76 824 accuracy 0.77 1707 macro avg 0.77 0.77 0.77 1707 weighted avg 0.77 0.77 0.77 1707