內(nèi)容簡(jiǎn)介：

本文使用python的sklearn庫(kù)對(duì)汽車歷史價(jià)格信息進(jìn)行回歸分析，包括數(shù)據(jù)預(yù)處理、特征相關(guān)性分析等步驟，最后利用lasso回歸建立價(jià)格預(yù)測(cè)模型。

數(shù)據(jù)集簡(jiǎn)介

數(shù)據(jù)中有分類變量也有連續(xù)變量，主要包括3類指標(biāo):
1. 汽車的各種特性.
2. 保險(xiǎn)風(fēng)險(xiǎn)評(píng)級(jí)：(-3, -2, -1, 0, 1, 2, 3).
3. 每輛保險(xiǎn)車輛年平均相對(duì)損失支付.

類別屬性 - make: 汽車的商標(biāo)（奧迪，寶馬。。。） - fuel-type: 汽油還是天然氣 - aspiration: 渦輪 - num-of-doors: 兩門還是四門 - body-style: 硬頂車、轎車、掀背車、敞篷車 - drive-wheels: 驅(qū)動(dòng)輪 - engine-location: 發(fā)動(dòng)機(jī)位置 - engine-type: 發(fā)動(dòng)機(jī)類型 - num-of-cylinders: 幾個(gè)氣缸 - fuel-system: 燃油系統(tǒng)連續(xù)指標(biāo) - bore: continuous from 2.54 to 3.94. - stroke: continuous from 2.07 to 4.17. - compression-ratio: continuous from 7 to 23. - horsepower: continuous from 48 to 288. - peak-rpm: continuous from 4150 to 6600. - city-mpg: continuous from 13 to 49. - highway-mpg: continuous from 16 to 54. - price: continuous from 5118 to 45400.

一、數(shù)據(jù)讀取與分析

# loading packages import numpy as np import pandas as pd from pandas import datetime# data visualization and missing values import matplotlib.pyplot as plt import seaborn as sns # advanced vizs import missingno as msno # missing values 對(duì)缺失值進(jìn)行可視化展示 %matplotlib inline# stats from statsmodels.distributions.empirical_distribution import ECDF from sklearn.metrics import mean_squared_error, r2_score# machine learning from sklearn.preprocessing import StandardScaler from sklearn.linear_model import Lasso, LassoCV from sklearn.model_selection import train_test_split, cross_val_score from sklearn.ensemble import RandomForestRegressor seed = 123# importing data ( ? = missing values) data = pd.read_csv("Auto-Data.csv", na_values = '?')# 查看數(shù)據(jù)信息 data.columns #查看每列列名 data.dtypes #查看字符類型 data.shape #查看數(shù)據(jù)有幾行幾列 data.head(5) #查看前面5行內(nèi)容

• 查看字符類型data.dtypes
  

• 查看數(shù)據(jù)形狀和前5行
  print("In total: ",data.shape)
  data.head(5)

• 數(shù)據(jù)集一共有, 26個(gè)特征，205行數(shù)據(jù)

• 數(shù)據(jù)描述性分析data.describe()
  

二、缺失值處理 （missingno缺失值可視化）

sns.set(style = "ticks") #指定風(fēng)格 msno.matrix(data) #畫圖

• 空白的地方表示存在缺失值，這里一共有7個(gè)特征數(shù)據(jù)包含缺失值，其中normalized-losses 缺失比較嚴(yán)重。

sns.set(style = "ticks") plt.figure(figsize = (12, 5)) c = '#366DE8'# ECDF plt.subplot(121) cdf = ECDF(data['normalized-losses']) #查看連續(xù)分布，累計(jì)結(jié)果 plt.plot(cdf.x, cdf.y, label = "statmodels", color = c); plt.xlabel('normalized losses'); plt.ylabel('ECDF');# overall distribution plt.subplot(122) plt.hist(data['normalized-losses'].dropna(), bins = int(np.sqrt(len(data['normalized-losses']))),color = c);

• 通過(guò)查看缺失值具體情況，可以發(fā)現(xiàn) 80% 的 normalized losses 是低于200 并且絕大多數(shù)低于125。
• 這種情況下，如果直接用中位數(shù)或者平均數(shù)來(lái)填充缺失值可能不夠精確。因此，可以考慮這個(gè)特征與那些因素有關(guān)系。

接下來(lái)，根據(jù)不同的風(fēng)險(xiǎn)等級(jí)來(lái)劃分組，在使用每組的平均數(shù)來(lái)填充normalized losses的缺失值。

#查看每組的情況 data.groupby('symboling')['normalized-losses'].describe()#刪除和填充缺失值 data = data.dropna(subset = ['price', 'bore', 'stroke', 'peak-rpm', 'horsepower', 'num-of-doors']) #對(duì)于缺失值少的幾列，直接刪掉缺失值 data['normalized-losses'] = data.groupby('symboling')['normalized-losses'].transform(lambda x: x.fillna(x.mean())) #填充缺失值#查看結(jié)果 print('In total:', data.shape) data.head()

• 缺失值處理之后，數(shù)據(jù)變成了193*26的數(shù)據(jù)集。

三、特征相關(guān)性

4.1 相關(guān)性計(jì)算和展示

cormatrix = data.corr() # cormatrix #查看結(jié)果#不同的展現(xiàn)格式 cormatrix *= np.tri(*cormatrix.values.shape, k=-1).T #返回函數(shù)的上三角矩陣，把對(duì)角線上的置0，讓他們不是最高的。 cormatrix = cormatrix.stack() #某一指標(biāo)與其他指標(biāo)的關(guān)系# 找出前十個(gè)最相關(guān)的特征 cormatrix = cormatrix.reindex(cormatrix.abs().sort_values(ascending=False).index).reset_index() cormatrix.columns = ["FirstVariable", "SecondVariable", "Correlation"] cormatrix.head(10)

• 第一行中，city_mpg 和 highway-mpg 兩個(gè)特征的相關(guān)性高達(dá)0.97，需要?jiǎng)h除掉其中一個(gè)。
• 對(duì)于數(shù)據(jù)中長(zhǎng)寬高，他們應(yīng)該存在某種配對(duì)關(guān)系，可以讓這幾個(gè)特征組合成一個(gè)新特征。

data['volume'] = data.length * data.width * data.heightdata.drop(['width', 'length', 'height', 'curb-weight', 'city-mpg'], axis = 1, # 1 for columnsinplace = True)

4.2 熱度圖展示

# Compute the correlation matrix corr_all = data.corr()# Generate a mask for the upper triangle mask = np.zeros_like(corr_all, dtype = np.bool) mask[np.triu_indices_from(mask)] = True# Set up the matplotlib figure f, ax = plt.subplots(figsize = (11, 9))# Draw the heatmap with the mask and correct aspect ratio sns.heatmap(corr_all, mask = mask,square = True, linewidths = .5, ax = ax, cmap = "BuPu") plt.show()

• 看起來(lái) price 跟這幾個(gè)的相關(guān)程度比較大 wheel-base,enginine-size, bore,horsepower.
  -也可以用seaborn展示具體的指標(biāo)情況 sns.pairplot(data, hue = 'fuel-type', palette = 'plasma')

4.3 進(jìn)一步回歸分析

從上面熱力圖中，得出價(jià)格price和另外幾個(gè)變量之間相關(guān)性很大。因此，接下來(lái)使用IMplot 進(jìn)一步查看兩個(gè)因素之間的關(guān)系。 lmplot對(duì)所選數(shù)據(jù)集進(jìn)行了一元線性回歸，擬合出了一條最佳的直線。

print('fuel_type:', data['fuel-type'].unique(), '\ndoors:', data['num-of-doors'].unique())

• fuel_type 和doors 是兩個(gè)分類變量，可根據(jù)兩個(gè)變量分組進(jìn)行分析。

sns.lmplot('price', 'horsepower', data, hue = 'fuel-type', col = 'fuel-type', row = 'num-of-doors', palette = 'plasma', fit_reg = True);

• 根據(jù)燃料的類型和門的數(shù)量的不同，劃分成4組，即4個(gè)圖。我們看到，無(wú)論是那種組合下，一輛汽車馬力與價(jià)格都是正相關(guān)。。

四、數(shù)據(jù)預(yù)處理

4.1 標(biāo)準(zhǔn)化

對(duì)連續(xù)值進(jìn)行標(biāo)準(zhǔn)化。

# target and features target = data.priceregressors = [x for x in data.columns if x not in ['price']] features = data.loc[:, regressors]num = ['symboling', 'normalized-losses', 'volume', 'horsepower', 'wheel-base','bore', 'stroke','compression-ratio', 'peak-rpm']# scale the data standard_scaler = StandardScaler() features[num] = standard_scaler.fit_transform(features[num])# glimpse features.head()

4.2 獨(dú)熱編碼

對(duì)分類屬性就行one-hot編碼。

# categorical vars classes = ['make', 'fuel-type', 'aspiration', 'num-of-doors', 'body-style', 'drive-wheels', 'engine-location','engine-type', 'num-of-cylinders', 'fuel-system']# create new dataset with only continios vars dummies = pd.get_dummies(features[classes]) features = features.join(dummies).drop(classes, axis = 1)# new dataset print('In total:', features.shape) features.head()

五、模型建立：Lasso回歸

5.1 劃分?jǐn)?shù)據(jù)集

# 按照30%劃分?jǐn)?shù)據(jù)集 X_train, X_test, y_train, y_test = train_test_split(features, target, test_size = 0.3, random_state = seed) print("Train", X_train.shape, "and test", X_test.shape)

-將數(shù)據(jù)集分成135的訓(xùn)練集和58的測(cè)試集。

5.2 Lasso回歸

基于線性回歸的基礎(chǔ)上，多加了一個(gè)絕對(duì)值想來(lái)懲罰過(guò)大的系數(shù)。

# logarithmic scale: log base 2 # high values to zero-out more variables alphas = 2. ** np.arange(2, 12) #指定alphas的范圍 scores = np.empty_like(alphas)for i, a in enumerate(alphas):lasso = Lasso(random_state = seed)lasso.set_params(alpha = a)lasso.fit(X_train, y_train)scores[i] = lasso.score(X_test, y_test)# 交叉驗(yàn)證cross validation lassocv = LassoCV(cv = 10, random_state = seed) lassocv.fit(features, target) lassocv_score = lassocv.score(features, target) lassocv_alpha = lassocv.alpha_plt.figure(figsize = (10, 4)) plt.plot(alphas, scores, '-ko') plt.axhline(lassocv_score, color = c) plt.xlabel(r'$\alpha$') plt.ylabel('CV Score') plt.xscale('log', basex = 2) sns.despine(offset = 15)print('CV results:', lassocv_score, lassocv_alpha)

5.3 特征重要性分析

# lassocv coefficients coefs = pd.Series(lassocv.coef_, index = features.columns)# prints out the number of picked/eliminated features print("Lasso picked " + str(sum(coefs != 0)) + " features and eliminated the other " + \str(sum(coefs == 0)) + " features.")# 展示前5個(gè)和后5個(gè) coefs = pd.concat([coefs.sort_values().head(5), coefs.sort_values().tail(5)])plt.figure(figsize = (10, 4)) coefs.plot(kind = "barh", color = c) plt.title("Coefficients in the Lasso Model") plt.show()

# 將 上面計(jì)算出來(lái)的Alphas 代入 model_l1 = LassoCV(alphas = alphas, cv = 10, random_state = seed).fit(X_train, y_train) y_pred_l1 = model_l1.predict(X_test)model_l1.score(X_test, y_test)

5.4 結(jié)果評(píng)估

5.4.1 residual plot 殘差圖

畫圖表示實(shí)際值和預(yù)測(cè)值之間的差異。

plt.rcParams['figure.figsize'] = (6.0, 6.0)preds = pd.DataFrame({"preds": model_l1.predict(X_train), "true": y_train}) preds["residuals"] = preds["true"] - preds["preds"] preds.plot(x = "preds", y = "residuals", kind = "scatter", color = c)

5.4.2 MSE和R2

#計(jì)算指標(biāo)：MSE和R2 def MSE(y_true,y_pred):mse = mean_squared_error(y_true, y_pred)print('MSE: %2.3f' % mse)return msedef R2(y_true,y_pred): r2 = r2_score(y_true, y_pred)print('R2: %2.3f' % r2) return r2MSE(y_test, y_pred_l1); R2(y_test, y_pred_l1);

5.4.3 查看具體實(shí)際值和預(yù)測(cè)值

#結(jié)果預(yù)測(cè) # predictions d = {'true' : list(y_test),'predicted' : pd.Series(y_pred_l1)}pd.DataFrame(d).head()

資料鏈接：https://edu.csdn.net/learn/7380/149737
missingno：https://github.com/ResidentMario/missingno

總結(jié)

以上是生活随笔為你收集整理的汽车价格预测回归分析模型的全部?jī)?nèi)容，希望文章能夠幫你解決所遇到的問(wèn)題。

如果覺(jué)得生活随笔網(wǎng)站內(nèi)容還不錯(cuò)，歡迎將生活随笔推薦給好友。