CELL-PHONE PRICE RANGE PREDICTION

BUSINESS CASE:

Find out some relation between features of a mobile phone(eg:- RAM, Internal Memory etc) and its selling price. In this problem you do not have to predict the actual price but a price range indicating how high the price is.

IMPORTING THE PYHTON LIBRARIES:

• NUMPY
• PANDAS
• MATPLOTLIB
• SEABORN

LOAD THE DATASET

DOMAIN ANALYSIS

INPUT VARIABLES :

• battery_power = Total energy a battery can store in one time measured in mAh(Continuous)
• blue = Has bluetooth or not (Categorical)
• clock_speed = speed at which microprocessor executes instructions(Continuous)
• dual_sim = Has dual sim support or not (Categorical)
• fc = Front Camera mega pixels(Continuous)
• four_g = Has 4G or not (Categorical)
• int_memory = Internal Memory in Gigabytes(Continuous)
• m_dep = Mobile Depth in cm(Continuous)
• mobile_wt = Weight of mobile phone(Continuous)
• n_cores = Number of cores of processor(Continuous)
• pc = Primary Camera mega pixels(Continuous)
• px_height = Pixel Resolution Height(Continuous)
• px_weight = Pixel Resolution Weight(Continuous)
• ram = Random Access Memory in Megabytes(Continuous)
• sc_h = Screen Height of mobile in cm(Continuous)
• sc_w = Screen Width of mobile in cm(Continuous)
• talk_time = longest time that a single battery charge will last when you are(Continuous)
• three_g = Has 3G or not (Categorical)
• touch_screen = Has touch screen or not (Categorical)
• wifi = Has wifi or not(Categorical)

OUTPUT VARIABLE

• Price Range = This is the target variable with value of 0(low cost), 1(medium cost), 2(high cost) and 3(very high cost).

  BASIC CHECKS

  • First 5 records of dataset
  • Last 5 records of dataset

  EXAMINE THE DATA

  • Data type of each column
  • Column names in dataset
  • Memory Usage
  • statistical analysis
  • CHECK THE CATEGORICAL COLUMNS
  • CHECK NORMALITY OF COLUMNS

EXPLORATORY DATA ANALYSIS

UNIVARIATE ANALYSIS

## INSIGHTS:

• None of the column is normally distributed.

• Distribution of fc,px_height,sc_w columns is left skewed.

• Distribution battery_power,clock_width,int_memory,m_dep,mobile_wt,pc,px_width,ram,sc_h,sc_w,talk_time columns has flat kurtosis.

  Check data balance

  

  INSIGHTS:

  • data is balanced

  BIVARIATE ANALYSIS

  Correlation of categorical variable with target

  INSIGHTS:

• Battery Power variable has correlation with target variable price range as battery power increases price of phone increases.

• internal memory variable has correlation with target variable price range as internal memory increases price of phone increases.

• px_width and px_height variable has correlation with target variable price range as px_width and px_height increases price of phone increases.

• ram has strong correlation with target variable as ram price of phone increases.

• primary camera also shows some corelation with target.

• 'clock_speed', 'fc','m_dep', 'mobile_wt', 'sc_h', 'sc_w', 'talk_time' does not have any corelation with increasing price of phone.

  

  INSIGHTS:

  • blue, dual_sim, n_cores, touch_screen, wifi has no correlation with target
  • three_g and four_g has correlation with target

  MULTIVARIATE ANALYSIS:

  

  DATA PRE-PROCESSING

  • CHECK NULL VALUES
  • CHECK OUTLIERS

HANDLING OUTLIERS

CHECKING CORRUPTED VALUES

• From EDA it is observed that,in the numerical columns there are 4 columns ('fc', 'pc', 'px_height', 'sc_w') whch have few 0 entries.
• However, the variables "front camera"(fc), "primary camera"(pc) having 0 as an entry can bes assumed that the mobile doesn't have front/rear camera.
• But the other two variables "pixel height"(pc_height) and "screen_width"(sc_w) can't have 0 as their values.
• Hence, these must be marked as corrupted.

SCALING

FEATURE ENGINEERING

ram, pixel_height, pixel_width, battery power, these are some featues which affects on cell phone price

clock_speed, m_dep, touch_screen, mobile_wt has very less corelation with target

TRAIN -TEST SPLIT

MODEL CREATION

1. LOGISTIC REGRESSION

EVALUATION Testing Accuracy 0.942 Training Accuracy 0.934375

precision    recall  f1-score   support

       0       0.99      1.00      1.00       105
       1       0.91      0.95      0.92        91
       2       0.92      0.84      0.88        92
       3       0.95      0.97      0.96       112

accuracy                           0.94       400
macro avg       0.94      0.94      0.94       400

HYPERPARAMETER TUNING

Testing Accuracy 0.975

precision    recall  f1-score   support

       0       1.00      0.96      0.98       105
       1       0.95      1.00      0.97        91
       2       0.98      0.96      0.97        92
       3       0.97      0.98      0.98       112

accuracy                           0.97       400
macro avg       0.97      0.98      0.97       400
weighted avg       0.98      0.97      0.98       400

2.SVM

EVALUATION

 precision    recall  f1-score   support

       0       0.95      0.94      0.95       105
       1       0.81      0.87      0.84        91
       2       0.75      0.78      0.77        92
       3       0.93      0.85      0.89       112

accuracy                           0.86       400
macro avg       0.86      0.86      0.86       400
weighted avg       0.87      0.86      0.86       400

MODEL COMPARISON

Model Accuracy:

• LogisticRegression 0.9750

• SVM 0.9625

• DecisionTree 0.8350

• RandomForest 0.8625

• BaggingClassifier 0.9725

• XGBoost 0.8950

FEATURE IMPORTANCE

USING LOGISTIC REGRESSION ALOGRITHM

RESULT: FOR ALL TOP ALGORITHMS ram, battery_power, px_height, px_width are strong features.