Machine learning practice 9-screening and predictive analysis of autism based on multiple models

Hello everyone, I am Wei Xue AI. Today I will introduce to you machine learning practice 9-screening and predictive analysis of autism based on multiple models. Autism is a neurodevelopmental disorder, mainly manifested in interpersonal communication and social interaction. Difficulty interacting, communication barriers, and repetitive stereotyped behaviors. Early screening and analysis are crucial for the diagnosis and intervention of children with autism.

Contents
1. Project background
2. Research significance
3. Code practice and data analysis
3.1 Data preprocessing
3.2 Data graph analysis
4. Machine learning model analysis
4.1 Data one-hot encoding
4.2 Data sorting
4.3 Logistic regression model
4.4 Random forest model
4.5 K nearest neighbor model
4.6 Running Results
5. Summary

1. Project background

Autism has received widespread attention over the past few decades, and its high prevalence and long-term impact on patients and their families is now recognized. However, due to the diverse symptoms of autism and the lack of specific biomarkers, its diagnosis and treatment face great challenges. Therefore, carrying out screening and analysis projects for autism can help improve the accuracy of early diagnosis and the effect of intervention.

2. Research Significance

Early Intervention: Early intervention for autism is critical to a child's development. Through screening and analysis programs, patients can be detected early and intervened in a timely manner, before children show obvious symptoms. This helps improve the patient's social interactions, language skills, and behavioral development.

Improve diagnostic accuracy: The diagnosis of autism relies on the clinical assessment of professional doctors, but this method has the risk of subjectivity and misdiagnosis. Through the screening and analysis project, advanced scientific technology and data analysis methods can be used to improve the diagnostic accuracy of autism and reduce missed and misdiagnosed cases.

Optimizing Resource Allocation: Diagnosis and treatment of autism requires significant time, financial and human resources. Through screening and analysis projects, we can better understand the epidemiological characteristics and social impact of autism, so as to optimize the allocation of resources and provide more effective support and services.

Promote research and knowledge accumulation: Screening and analysis projects can collect a large amount of data, providing valuable resources and information for autism research. This will help to gain insight into the pathogenesis, genetic factors and potential treatments of autism, and promote scientific progress in the field of autism.

3. Code combat and data analysis

3.1 Data preprocessing

First of all, you need to load the data set, the download address of the data set:
Link: https://pan.baidu.com/s/1sfb3_w2o5X7ya7Z0R51Npw?pwd=94we
Extraction code: 94we

# 第三方库导入
import numpy as np # 导入numpy库用于进行线性代数计算
import pandas as pd # 导入pandas库用于数据处理
import matplotlib.pyplot as plt # 导入matplotlib库用于数据可视化
import seaborn as sns # 导入seaborn库用于数据可视化
# 读取数据集1和数据集2
df1 = pd.read_csv('Autism_Data.arff', na_values='?')
df2 = pd.read_csv('Toddler Autism dataset July 2018.csv', na_values='?')

sns.set_style('whitegrid') # 设置seaborn风格为白色网格

# 提取ASD类别为YES的数据（成年人）
data1 = df1[df1['Class/ASD'] == 'YES']

# 提取ASD Traits为Yes的数据（幼儿）
data2 = df2[df2['Class/ASD Traits '] == 'Yes']

# 计算ASD阳性成年人的比例
print("成年人: ", len(data1) / len(df1) * 100)

# 计算ASD阳性幼儿的比例
print("幼儿:", len(data2) / len(df2) * 100)

# 创建一个包含2个子图的画布，设置大小为20x6
fig, ax = plt.subplots(1, 2, figsize=(20, 6))

3.2 Data graphic analysis

Heatmap of missing values ​​for an adult dataset

sns.heatmap(data1.isnull(), yticklabels=False, cbar=False, cmap='viridis', ax=ax[0])
ax[0].set_title('成年人数据集')
ax[0].set_ylabel('样本索引')

Heatmap missing values ​​for a toddler dataset

sns.heatmap(data2.isnull(), yticklabels=False, cbar=False, cmap='viridis', ax=ax[1])
ax[1].set_title('幼儿数据集')
ax[1].set_ylabel('样本索引')

plt.show() # 显示图形

Draw a count histogram of jaundice at birth in adults and children with ASD positive

# 创建一个包含2个子图的画布，设置大小为20x6
fig, ax = plt.subplots(1, 2, figsize=(20, 6))

# 绘制成年人ASD阳性中出生时黄疸情况的计数柱状图
sns.countplot(x='jundice', data=data1, hue='gender', ax=ax[0])
ax[0].set_title('成年人ASD阳性中出生时黄疸情况的性别分布')
ax[0].set_xlabel('出生时黄疸情况')

# 绘制幼儿ASD阳性中出生时黄疸情况的计数柱状图
sns.countplot(x='Jaundice', data=data2, hue='Sex', ax=ax[1])
ax[1].set_title('幼儿ASD阳性中出生时黄疸情况的性别分布')
ax[1].set_xlabel('出生时黄疸情况')

plt.show() # 显示图形

Plot a histogram of the age distribution of ASD positivity in adults, young children

# 创建一个包含2个子图的画布，设置大小为20x6
fig, ax = plt.subplots(1, 2, figsize=(20, 6))

# 绘制成年人ASD阳性年龄分布的直方图
sns.distplot(data1['age'], kde=False, bins=45, color='darkred', ax=ax[0])
ax[0].set_xlabel('年龄（岁）')
ax[0].set_title('ASD阳性成年人年龄分布')

# 绘制幼儿ASD阳性年龄分布的直方图
sns.distplot(data2['Age_Mons'], kde=False, bins=30, color='darkred', ax=ax[1])
ax[1].set_xlabel('年龄（月）')
ax[1].set_title('ASD阳性幼儿年龄分布')

plt.show() # 显示图形

Analysis of mapping the country distribution of adults with positive ASD

plt.figure(figsize=(20,6))
sns.countplot(x='contry_of_res',data=data1,order= data1['contry_of_res'].value_counts().index[:15],hue='gender',palette='viridis')
plt.title('Positive ASD Adults country wise distribution')
plt.xlabel('Countries')
plt.tight_layout()
plt.show() # 显示图形
# 输出种族的计数值
print(data1['ethnicity'].value_counts())
data2['Ethnicity'].value_counts()

# 绘制白人和欧洲人种族在各个国家的分布图
plt.figure(figsize=(15,6))
sns.countplot(x='contry_of_res',data=data1[data1['ethnicity']=='White-European'],order=data1[data1['ethnicity']=='White-European']['contry_of_res'].value_counts().index[:10],palette='viridis')
plt.title('Positive ASD of White and European Ethnicities country wise distribution')
plt.xlabel('Countries')
plt.tight_layout()
plt.show() # 显示图形
# 绘制不同种族的 ASD 成人亲属中有无自闭症分布和不同种族的 ASD 儿童亲属中有无自闭症分布
fig, ax = plt.subplots(1,2,figsize=(20,6))
sns.countplot(x='austim',data=data1,hue='ethnicity',palette='rainbow',ax=ax[0])
ax[0].set_title('Positive ASD Adult relatives with Autism distribution for different ethnicities')
ax[0].set_xlabel('Adult Relatives with ASD')
sns.countplot(x='Family_mem_with_ASD',data=data2,hue='Ethnicity',palette='rainbow',ax=ax[1])
ax[1].set_title('Positive ASD Toddler relatives with Autism distribution for different ethnicities')
ax[1].set_xlabel('Toddler Relatives with ASD')
plt.tight_layout()

4. Machine learning model analysis

4.1 Data one-hot encoding

within24_36= pd.get_dummies(df2['Age_Mons']>24,drop_first=True)  # 大于24个月的为1，否则为0
within0_12 = pd.get_dummies(df2['Age_Mons']<13,drop_first=True)  # 小于13个月的为1，否则为0
male=pd.get_dummies(df2['Sex'],drop_first=True)  # 性别为男性的为1，否则为0
ethnics=pd.get_dummies(df2['Ethnicity'],drop_first=True)  # 使用独热编码表示种族
jaundice=pd.get_dummies(df2['Jaundice'],drop_first=True)  # 是否有黄疸，有黄疸为1，否则为0
ASD_genes=pd.get_dummies(df2['Family_mem_with_ASD'],drop_first=True)  # 亲属中是否有自闭症，有自闭症为1，否则为0
ASD_traits=pd.get_dummies(df2['Class/ASD Traits '],drop_first=True)  # ASD 特征，有特征为1，否则为0

4.2 Data collation

import pandas as pd

# 将多个数据集按列合并
final_data = pd.concat([within0_12, within24_36, male, ethnics, jaundice, ASD_genes, ASD_traits], axis=1)

# 设置列名
final_data.columns = ['within0_12', 'within24_36', 'male', 'Latino', 'Native Indian', 'Others', 'Pacifica', 'White European', 'asian', 'black', 'middle eastern', 'mixed', 'south asian', 'jaundice', 'ASD_genes', 'ASD_traits']

# 显示合并后的数据的前几行
final_data.head()

from sklearn.model_selection import train_test_split
# 划分特征和标签
X = final_data.iloc[:, :-1]
y = final_data.iloc[:, -1]

# 划分训练集和测试集
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=101)

4.3 Logistic regression model

from sklearn.linear_model import LogisticRegression

# 创建逻辑回归模型
logmodel = LogisticRegression()

# 在训练集上训练逻辑回归模型
logmodel.fit(X_train, y_train)

from sklearn.model_selection import GridSearchCV

# 设置网格搜索的参数
param_grid = {
    
    'C': [0.01, 0.1, 1, 10, 100, 1000]}

# 创建逻辑回归模型的网格搜索对象
grid_log = GridSearchCV(LogisticRegression(), param_grid, refit=True)

# 在训练集上进行网格搜索
grid_log.fit(X_train, y_train)

print('GridSearchCV')
# 输出网格搜索得到的最佳模型参数
print(grid_log.best_estimator_)

# 使用网格搜索得到的最佳模型在测试集上进行预测
pred_log = grid_log.predict(X_test)

from sklearn.metrics import classification_report, confusion_matrix

# 输出逻辑回归模型在测试集上的混淆矩阵和分类报告
print(confusion_matrix(y_test, pred_log))
print(classification_report(y_test, pred_log))

4.4 Random Forest Model

from sklearn.ensemble import RandomForestClassifier

# 创建随机森林分类器
rfc = RandomForestClassifier(n_estimators=100)

# 在训练集上训练随机森林分类器
rfc.fit(X_train, y_train)

# 使用随机森林分类器在测试集上进行预测
pred_rfc = rfc.predict(X_test)

print('RandomForestClassifier')
# 输出随机森林分类器在测试集上的混淆矩阵和分类报告
print(confusion_matrix(y_test, pred_rfc))
print(classification_report(y_test, pred_rfc))

4.5 K nearest neighbor model

from sklearn.preprocessing import StandardScaler

# 对特征进行标准化处理
scaler = StandardScaler()
scaler.fit(X)
scaled_features = scaler.transform(X)
X_scaled = pd.DataFrame(scaled_features, columns=X.columns)

X_train, X_test, y_train, y_test = train_test_split(X_scaled, y, test_size=0.2, random_state=101)

from sklearn.neighbors import KNeighborsClassifier

# 计算不同的K值下的分类错误率
error_rate = []
for i in range(1, 50):
    knn = KNeighborsClassifier(n_neighbors=i)
    knn.fit(X_train, y_train)
    pred_i = knn.predict(X_test)
    error_rate.append(np.mean(pred_i != y_test))

# 绘制K值和错误率的关系图
plt.figure(figsize=(10, 6))
plt.plot(range(1, 50), error_rate, color='blue', linestyle='dashed', marker='o', markerfacecolor='red', markersize=10)
plt.title('Error rate vs K-value')
plt.xlabel('K')
plt.ylabel('Error Rate')

# 根据错误率最低的K值创建K近邻分类器
knn = KNeighborsClassifier(n_neighbors=13)
knn.fit(X_train, y_train)

# 使用K近邻分类器在测试集上进行预测
pred_knn = knn.predict(X_test)

print(confusion_matrix(y_test, pred_knn))
print(classification_report(y_test, pred_knn))

4.6 Running Results

Logistic regression model:

       precision    recall  f1-score   support

           0       0.00      0.00      0.00        78
           1       0.63      1.00      0.77       133

    accuracy                           0.63       211
   macro avg       0.32      0.50      0.39       211
weighted avg       0.40      0.63      0.49       211

Random forest model:

          precision    recall  f1-score   support

           0       0.71      0.37      0.49        78
           1       0.71      0.91      0.80       133

    accuracy                           0.71       211
   macro avg       0.71      0.64      0.64       211
weighted avg       0.71      0.71      0.68       211

K nearest neighbor classification model:

            precision    recall  f1-score   support

           0       0.68      0.32      0.43        78
           1       0.70      0.91      0.79       133

    accuracy                           0.69       211
   macro avg       0.69      0.62      0.61       211
weighted avg       0.69      0.69      0.66       211

5. Summary

This article analyzes the situation of autism through the Toddler Autism dataset July 2018.csv dataset, and performs visual analysis through code and charts, where functions are used to pd.concat()combine multiple datasets into one final_datadataset by column. Then separate the features and labels and use train_test_split()a function to split the data into training and testing sets.

In this paper, the grid search logistic regression model, random forest model and K-nearest neighbor classifier are used to train the training set and make predictions on the test set. Finally, output the confusion matrix and classification report of the model to evaluate the model performance.

Among them, after the features are standardized, the K value is used to search in the range from 1 to 49, and the K value with the lowest error rate is found, and the final K-nearest neighbor classifier is created for prediction and evaluation.