A Framework For Airfare Price Prediction: A Machine .
A Framework for Airfare Price Prediction: A Machine Learning ApproachTianyi Wang , Samira Pouyanfar , Haiman Tian , Yudong Tao† ,Miguel Alonso Jr. , Steven Luis and Shu-Ching Chen School of Computing and Information SciencesFlorida International University, Miami, Florida 33199Emails: cs.fiu.edu† Department of Electrical and Computer EngineeringUniversity of Miami, Coral Gables, FL 33124Email: yxt128@miami.eduAbstract—The price of an airline ticket is affected by anumber of factors, such as flight distance, purchasing time,fuel price, etc. Each carrier has its own proprietary rulesand algorithms to set the price accordingly. Recent advance inArtificial Intelligence (AI) and Machine Learning (ML) makesit possible to infer such rules and model the price variation.This paper proposes a novel application based on two publicdata sources in the domain of air transportation: the AirlineOrigin and Destination Survey (DB1B) and the Air CarrierStatistics database (T-100). The proposed framework combinesthe two databases, together with macroeconomic data, and usesmachine learning algorithms to model the quarterly averageticket price based on different origin and destination pairs, asknown as the market segment. The framework achieves a highprediction accuracy with 0.869 adjusted R squared score onthe testing dataset.Keywords-machine learning; airfare price; DB1B; T-100; prediction model;I. I NTRODUCTIONSince the deregulation of the airline industry, airfarepricing strategy has developed into a complex structureof sophisticated rules and mathematical models that drivethe pricing strategies of airfare [1] [2] [3]. Although stilllargely held in secret, studies have found that these rules arewidely known to be affected by a variety of factors [4] [5].Traditional variables such as distance, although still playinga significant role, are no longer the sole factor that dictate thepricing strategy. Elements related to economic, marketingand societal trends have played increasing roles in dictatingthe airfare prices.Most studies on airfare price prediction have focused oneither the national level or a specific market. Research atthe market segment level, however, is still very limited. Wedefine the term market segment as the market/airport pairbetween the flight origin and the destination. Being able topredict the airfare trend at the specific market segment levelis crucial for airlines to adjust strategy and resources for aspecific route. However, existing studies on market segmentprice prediction use heuristic-based conventional statisticalmodels, such as linear regression [6] [7], and are based onthe assumption that there exists a linear relationship betweenthe dependent and independent variables, which in manycases, may not be true.Recent advances in Artificial Intelligence (AI) and Machine Learning (ML) make it possible to infer rules andmodel variations on airfare price based on a large numberof features, often uncovering hidden relationships amongstthe features automatically. To the best of our knowledge, allexisting work leveraging machine learning approaches forairfare price prediction are based on: 1) proprietary datasetsthat are not publicly available [8] [9] and 2) transactionrecords data crawled from online travel booking sites likeKayak.com [10] [11] [12]. The problem of the former liesin the difficulty of gaining access to the data, making reproducing the results and extending the work nearly impossible.The issue with the later is that the transaction records fromeach online booking site are a small fraction of the totalticket sales from the entire market, making the acquired datalikely to be skewed, and thus, not representing the true natureof the entire market.In this paper, we address the problem of market segmentlevel airfare price prediction by using publicly availabledatasets and a novel machine learning framework to predictmarket segment level airfare price. More specifically, ourproposed framework extracts information from two specificpublic datasets, the DB1B and the T-100 datasets that arecollected and maintained by the Office of Airline Information within the United States Bureau of TransportationStatistics (BTS). The DB1B dataset has been utilized invarious studies that assess the determinants of aircraft characteristics and frequency of flights [13], analyses for thestructure and dynamics of O-D for the core of the airtravel market [14], and demand-prediction [15]. The T-100dataset includes air passenger volumes for U.S. domestic andinternational markets and covers large certified carriers thathold Certificates of Public Convenience and Necessity. Thegoal of our proposed framework is to draw a comprehensiveprofile of each market and uses machine learning techniquesto predict the average airfare on market segment level.The remainder of this paper is organized as follows. Section II reviews existing work that utilized either conventional
statistical or machine learning algorithms for airfare priceprediction. Section III provides a detailed description ofthe two datasets and the proposed framework. Section IVdescribes the experimental setup and presents the results ofapplying our proposed framework, as well as a comparisonwith several baseline methods. In section V, we concludethe paper with a discussion of our contribution and severalpotential directions for future work.II. R ELATED W ORKAir ticket price prediction is a challenging task sincethe factors involved in pricing dynamically change overtime and make the price fluctuate. In the last decade,researchers have incorporated machine learning algorithmsand data mining strategies to better model observed prices.Among them, regression models, such as Linear Regression(LR), Support Vector Machines (SVMs), Random Forests(RF), are frequently used in predicting accurate airfareprice [10][16][17].Early work also considered using classification models topredict the trends of the itineraries. Ren et al. [17] proposedusing LR, Naive Bayes, Softmax regression, and SVMs tobuild a prediction model and classify the ticket price into fivebins (60% to 80%, 80% to 100%, 100% to 120%, and etc.)to compare the relative values with the overall average price.More than nine thousand data points, including six features(e.g., the departure week begin, price quote date, the numberof stops in the itinerary, etc.), were used to build the models.The authors reported the best training error rate close to22.9% using LR model. Their SVM regression model failedto produce a satisfying result. Instead, an SVM classificationmodel was used to classify the prices into either “higher” or“lower” than the average.In [16], four LR models were compared to obtain the bestfit model, which aims to provide an unbiased information tothe passenger whether to buy the ticket or wait longer fora better price. The authors suggested using linear quantilemixed models to predict the lowest ticket prices, which arecalled the “real bargains”. However, this work is limitedto only one class of tickets, economy, and only on onedirection single leg flights from San Francisco Airport toJohn F. Kennedy Airport. Wohlfarth et al. [18] integratedclustering as a preliminary stage with multiple state-ofthe-art supervised learning algorithms (classification tree(CART) and RF) to assist the customers’ decision makingprocess. Their framework uses the K-Means algorithm togroup flights with similar behavior in the price series. Theythen use CART to interpret meaningful rules, and RF toprovide information about the importance of each feature.Also, the authors pointed out that one element, the numberof seats left, is a key freature for ticket price prediction.Aside from flight-specific features, many other attributesaffect the competitive market. Accurately predicting themarket demand, for example, can reduce a travel agency’saccumulated costs, which are caused by over purchasingor lost orders. In [19], the author applied Artificial NeuralNetwork (ANN) and Genetic Algorithms (GA) to predictair ticket sales revenue for the travel agency. The input features included international oil price, Taiwan stock marketweighted index, Taiwan’s monthly unemployment rate, andso on. Specifically, the GA selects the optimum input features to improve the performance of the ANNs. The modelshowed good performance with a 9.11% Mean AbsolutePercentage Error.Starting from 2017, more advanced machine learningmodels have been considered to improve airfare price prediction [10][20]. Tziridis et al. [10] applied eight machinelearning models, which included ANNs, RF, SVM, and LR,to predict tickets prices and compared their performance.The best regression model achieved an accuracy of 88%.In their comparison, Bagging Regression Tree is identifiedas the best model, which is robust and not affected byusing different input feature sets. In [20], Deep RegressorStacking was proposed to reach more accurate predictions.The proposed method is a novel multi-target approach withRF and SVM as the regressors and can be easily applied toother similar problem domains.As airline ticket data is not well organized and ready fordirect analysis, collecting and processing those data alwaysrequires a great deal of effort. For most analyses found in theliterature, researchers evaluate their models’ performance ondifferent datasets by either crawling the data from the web orrequesting private data from collaborative organizations. Asa result, it is difficult to replicate the research and conductcomparisons of the models’ performance. For U.S. airlines,the fare data is publicly available in the T100 and DB1A/1Bdatabases. However, due to the limited association betweenthe prices and specific flights information, these datasets areseldom used independently to generate scientific researchoutcomes [21]. However, researchers who are interestedin analyzing the price dispersion, for example, are morelikely to consider investigating the information from thosedatasets [22]. In Rama-Murthy’s dissertation [7], the OfficialAirline Guide (OAG) and DB1B data are used to modelthe airfare prices. The author also incorporates the SabreAirPrice data, which was provided by SABRE, but they onlyprovide the information of their online users. As this onlineuser data does not represent the whole consumer market, itcan bias the results obtained from the data.Compared to the current and recent work, our proposedframework manages to handle the price prediction task onlyusing public data sources with minimal features. Also, notrestricted by any specific market segment that usually limitsthe existing work, this proposed framework can be appliedto predict the airfare price for any market.
Table IS UMMARY OF DATA IN DB1B AND T-100 USED IN THE PROPOSED FRAMEWORKEntityTicketCouponMarket segmentMarket segment by carrierAvailabilityDB1BDB1BDB1B&T-100T-100Datafare price, total distance, and total number of passengersmarket segment, time of the itinerary, carrier, and seat classoriginal airport, destination airport, and segment distancenumber of passengers, and number of available seats by aircraft typeIII. M ATERIALS AND M ETHODA. DatasetsIn order to build the airline ticket price model at themarket segment level, information about both the airlinetraffic and passenger volume for each market segment isrequired. Therefore, two public datasets (DB1B and T-100)are used in our proposed framework. Data collected during2018 are used to train and evaluate the proposed model.Table I summarizes the information of these two datasets.The United States Department of Transportation regularlyupdates both the DB1B and the T-100. The DB1B datasetprovides quarterly-aggregated information about the airlinetickets in the United States from reporting carriers andconsists of 10% randomly sampled ticket data from eachreporting carrier. The information in DB1B is organized inthree parts, namely “Coupon”, “Market”, and “Ticket”. A“Coupon” is an atomic unit of an airline ticket, indicatingone itinerary of a passenger that is directly transportedfrom one airport to another, while each ticket could contain multiple coupons and multiple passengers. Therefore,“Coupon” in the DB1B includes information about each leg,“Market” provides information on the market segment, suchas the distance between two airports, and “Ticket” providesadditional information at ticket level, such as airfare price.All the records in “Coupon” are bounded to a “Market”record and a “Ticket” record. For our proposed framework,a subset of most related data are used, including the originairport (ORIGIN), the destination airport (DEST), time ofthe itinerary (QUARTER), carrier information (REPORTING CARRIER), seat class (SEAT CLASS) (e.g., first,business, economic, etc.), total flight distance for a ticket(DISTANCE), airfare price (ITIN FARE), and the numberof passengers in a ticket (PAX).Different from DB1B, T-100 provides monthly domesticnon-stop segment data reported by both the domestic andinternational carriers. It presents the number of passengersof each airline and each market segment by aircraft type.B. Proposed FrameworkOur proposed framework utilizes both the DB1B and T100 datasets, in combination with macroeconomic data topredict the quarterly average airfare at the market segmentlevel. Figure 1 shows an overview of the major components of the proposed framework. In the data preprocessingstep, all datasets are cleaned to exclude possible erroneousAirline Origin andDestinationSurvey (DB1B)DataPreprocessingAir Carrier StatisticsDatabase (T-100)Data PreprocessingFeature ExtractionFeature SelectionAirfareEconomic DataPrediction ModelFigure 1. Proposed framework for airfare price prediction using publicdata sourcessamples, transformed and combined based on the marketsegment. The feature extraction module serves to extract andgenerate handcrafted features that aim to characterize themarket segment. The goal of the feature selection moduleis to optimize the performance of the prediction model byanalyzing the effectiveness of the features and remove anyirrelevant features. Finally, we use the selected features tobuild our prediction model, which generates the output valueas the predicted air ticket price.1) Data Preprocessing: In the DB1B and T-100 datasets,many attributes contain the same information. Directly merging the tables creates many duplicate fields. Also, the datareported by the airlines may include erroneous values causedby human error, currency conversion error, etc. Therefore, aproperly designed data preprocessing workflow is crucial togenerate accurate input data in order to build the machinelearning model.Table II shows the layout of the ticket and coupondatabase tables and the sample records with the sameitinerary ID (ITIN ID) in the DB1B dataset. First, theDB1B ticket and coupon tables are merged based on theITIN ID. The ITIN ID is the primary key for the tickettable. In the coupon table, all of the entries that belong tothe same ticket share the same ITIN ID. Samples in theDB1B ticket table with the itinerary value (ITIN FARE) lessthan 50, or distance field (DISTANCE) less than 100 milesin the Coupon table are removed because those samples inpractice, are considered reporting errors. Samples with pricecredibility field (DOLLAR CRED) equal to 0 are unreliable
Table IIS TRUCTURE OF THE DB1B TICKET AND COUPON TABLE WITH SAMPLE RECORDSITIN ID2018112.QUARTER1.ORIGINABE.ITIN FARE340.ITIN .DB1B Ticket TableDISTANCE1384.DB1B Coupon TableREPORTING CARRIER9E9E.Table IIIS TRUCTURE OF THE T-100 DATASET WITH A SAMPLE QUARTER1.carrier reports, which are also disregarded. Since only theticket table contains the ticket price, the price for eachmarket segment is calculated based on the ITIN FARE inthe ticket table and the distance ratio. The distance ratiomeasures the proportion between the distance of each leg inthe coupon table and the full length of the itinerary in theticket table. Finally, the quarterly average fare value for eachSEAT CLASS on each specific market segment is generated.Table III shows an example record in the T-100 database.Similar to the DB1B, the “SEATS” and “PAX” fields inT-100 are aggregated based on the origin and destinationairports pair for each quarter. In the final stage, the two datasources consisting of the cleaned attributes are merged basedon the market segment and on a quarterly basis.2) Feature Extraction: Several features have been extracted from the DB1B and T-100 dataset to representa specific aspect of the market segment. Furthermore, toexploit the relationship between the airline industry andthe overall economic conditions, several macroeconomicfeatures are also added to the feature set. Table IV describesall the features that are identified during feature extraction.Table IVT HE LIST OF FEATURES GENERATED DURING FEATURE EXTRACTIONSTAGE WITH EXPLANATIONSFeature NameDistanceSeat ClassPassenger VolumeLoad FactorCompetition FactorLCC PresenceCrude Oil PriceCPIQuarterDescriptionMarket distance between the origin anddestination airportsIndicator for economy or premiumseat typeTotal number of passengers traveledbetween the origin and destination airportsThe ratio of the total number of passengerto the total number of seat in a marketThe market HHIIndicator of LCC operating in the marketQuarterly average of crude oil priceQuarterly average of Consumer Price IndexIndicates the three month period of the yearPAX1.DOLLAR CRED1.PAX11.SEAT CLASSXX.DISTANCE692692.The Load Factor (LF) is a primary metric used in thetransportation industry. It represents the supply and demandrelationship in a given market, which strongly influencesan airline’s pricing strategy. The T-100 dataset includes twofeatures, the number of available seats and the number ofactual passengers carried, that allow us to calculate the LFof a market by dividing the total passenger volume (P ) bythe total number of Available Seats (AS) in that marketsegment:P(1)LF ASThe effect of competition among airlines in a given marketsegment has been shown to affect the pricing strategy of theairlines [23]. In a less competitive market, the market powerof a given airline is stronger, and thus, it is more likely toengage in price discrimination. On the other hand, the higherthe level of competition, the weaker of the market power ofan airline, and then the less likely the chance of the airlinefare increases. The competition factor in the proposed modelis based on the Herfindahl-Hirschman Index (HHI) [24],which measures the level of competition in a given market.It is the sum of the squared fraction of the market share ofeach top company:HHI CXsa ,(2)a 1va,(3)Pwhere C is the total number of companies, sa is the marketshare of company a, v a is the number of passenger carriedby company a, and P is the total number of passengerin the market. We used the T-100 dataset to extract themarket share of each airline in a specific market segmentby calculating the ratio of the number of passengers carriedby that airline to the total passenger volume of the marketsegment.The emergence of Low-Cost Carrier (LCC) has revolutionized the entire operating model of the airline industry.The presence of LCC in a market has had a substantialimpact on the total passenger volume and the air ticketprice [25]. A “LCC Presence” field is added to indicatewhether the “Carrier” field in the DB1B coupon tablesa
contains the International Air Transport Association (IATA)code [26] related to one of the LCCs operating in the U.S.domestic markets. The six LCCs are Allegiant Air, FrontierAirlines, JetBlue, Southwest Airlines, Spirit Airlines, andSun Country Airlines.Macroeconomic data, such as crude oil price and Consumer Price Index (CPI), can also be utilized to uncoverthe hidden trend in airline fares. Fuel costs can take up to50% of the total operating cost of an airline [27]. Hence, thelevel of crude oil price plays an essential rule of formulatingthe airline’s pricing strategy. It is a common practice forairlines to pass the cost of aviation fuel to the customerby adjusting the fare to compensate for the fluctuation ofcrude oil price. In this paper, we used the West TexasIntermediate (WTI) crude oil price data and calculated itsquarterly average value. Furthermore, the CPI measures theweighted average prices of various types of consumer goodsand services, which include the prices in the transportationindustry [28]. Therefore, we exploit its potential to measurethe current level of air travel cost. The monthly CPI data isacquired from the Organization for Economic Co-operationand Development. Similar to the crude oil price, we calculatethe quarterly averaged value. Figure 2 depicts the quarterlyvalue trend of crude oil price, CPI, and airfare from 2006 to2017. It demonstrates a clear relationship between the threetypes of data.by their importance. A feature’s importance is measured bythe average decrease in impurity. It is the total decrease inthe node’s impurity caused by the corresponding feature,weighted by the chance that the decision path includesthis node. There are several ways of choosing the impuritymetric, and because our target is a continuous value, Sumof Squared Errors (SSE) is chosen as the impurity metric.The SSE for node o can be calculated as:SSEo SX 2j ,(4)j 1where S is the number of splits from the node, and is theerror between the true value and predicted value. The NodeImportance (NI) for node o can be calculated as (assumingthe parent node splits into two child nodes):N Io wo SSEo wl SSEl wr SSEr ,(5)where wo , wl and wr are the weighted number of samplespass through node o and it’s left and right child node. Then,the Feature Importance (FI) for feature x can be calculatedasPb,b nodes split on f eature x N IbP.(6)F Ix k,k all nodes N IkGenerally, a feature gains more importance when it has agreater effect of reducing the prediction error.In the next step, the feature selection module applies Recursive Elimination (RE) to select the best set of features forthe prediction model. More specifically, for each iteration,the feature with the lowest feature importance is eliminated,and the model will be retrained on the updated input. Thisprocess terminates when the act of removing more featuredoes not improve the model’s performance.4) Machine learning model: When developing the MLmodel, we chose RF as the learner for the airfare priceprediction task. Based on our empirical study, the RF modeldemonstrates the best performance on the data as comparedto several ML techniques including LR, SVM, and neuralnetworks. Comparison results are explained in Section IV.IV. E XPERIMENTS AND ANALYSISFigure 2. A comparison between the crude oil price, CPI and the quarterlyaveraged airfare from 2006 to 20173) Feature Selection: A feature selection technique isapplied to improve the model performance by investigatingthe degree of impact of each feature on the prediction result.We utilize RF to construct an automated feature selectionmodule. RF is a tree-based ensemble learning algorithm thatbuilds multiple decision tree classifier during the trainingphase and outputs the predicted results based on either themajority vote (classification) or the average (regression) ofthe predictions of all decision trees. After training the RFmodel with the entire feature set, it ranks all the featuresA. Experimental SetupFor our experiments, we collected 16,577,497 and41,360,566 samples from the 2018 DB1B ticket tableand coupon table, respectively. The T-100 dataset contains329,426 samples. We tested several well-known machinelearning models as baselines to compare with the RF model.In particular, LR, SVM, Multilayer Perceptrons (MLPs), andXGBoost Tree were used for the evaluation. For the SVMmodel, the radial basis function kernel is used, the tolerancefor stopping criterion is set to 0.001, the penalty parameterfor the error term is set to 0.1. For the MLPs, three hiddenlayers are used with 30 neurons per layer. The RectifiedLinear Unit (ReLU) [29] is used as the activation function
0.6Importance 0.01930.0Figure 3.vuNu1 X(yi yˆi )2RM SE tN i 1(7)where N is the total sample size. The lower the RMSE valueis, the higher performance the regression model has.2The Adjusted R Squared, (Radj), is usually used to explainhow well the independent variables fit a curve or line.Adjusted R2 also adjusts for the number of variables in amodel. The higher the Adjusted R Squared is, the better theresult of regression is. It is calculated as follows: (1 R2 )(N 1)N p 1 (8)where p is the number of predictors and R2 is:PN(yi yˆi )2R2 1 Pi 1N2i 1 (yi ȳ)here ȳ is the mean value of y.0.00690.0049Importance score value for each feature generated by RFand Adam is the optimization function [30]. The learningrate is set to 0.0001 with momentum enabled set to 0.9.For the XGBoost model, the number of estimators is set to100 with a learning rate as 0.1, and max depth equals to5. For the RF model, the number of estimators is also setto 100 with the minimum number of samples to split setto 2. To evaluate the proposed price prediction model, twopopular metrics for regression analysis are used: the RootMean Square Error (RMSE) and the Adjusted R Squared.RMSE calculates the differences between the observedvalues, y, and predicted values, ŷ. This difference for eachdata point is also called the residual. Thus, RMSE iscalculated as follows:2Radj 1 0.0083(9)Table VP ERFORMANCE COMPARISON FOR DIFFERENT REGRESSION MODELSWITH AND WITHOUT FEATURE SELECTIONMethodLRSVMMLPXGBoostRFwithout feature 75483.4810.77866.5840.858with onR2adj0.6180.6260.7660.7970.869B. Experimental ResultsIn order to demonstrate the importance of each featurefor airfare price prediction, we extracted the importancescores generated by the feature selection module. Figure 3depicts the importance value for each feature. As shownin the figure, “Distance” and “Seat Class” (Economy orbusiness) are the most important factors for airfare priceestimation followed by “Passenger Volume”, “Load factor”,and “Competition Factor”. Although the “CPI” and “CrudeOil Price” do not have very high importance scores, theycan still help the model predict a more accurate estimationof the airfare price. However, based on our experiments,“Quarter” does not help the regression model. Includingthe variable “Quarter” does not reduce the error during thetraining phase. Thus, it is automatically removed by the RFfeature selection module. The goal is to identify the featuresthat improve the model’s performance and adding irrelevantfeatures deteriorates the model’s performance, as the modellearns an irrelevant pattern.The results comparing various regression models withfeature selection and without feature selection are shown inTable V. As can be seen from this table, LR and SVM havethe lowest performance compared to other ML methods with2respect to the RMSE and Radjmetrics. The performance ofall of the models improves after applying feature selection,
Table VIP ERFORMANCE COMPARISON FOR DIFFERENT REGRESSION MODELSWITHOUT L OAD FACTOR , C OMPETITION FACTOR , CPI, AND C RUDEO IL P RICE ch illustrates the importance of this module. XGBoostperforms better than MLP, SVM, and LR, but does notoutperform RF for airfare price prediction. Therefore, weutilize RF in the proposed framework, which achieves thehighest performance compared to other baselines for thisdataset. Specifically, it reaches 62.753 and 0.869 RMSE and22Radj, respectively. In other words, it improves the Radjby40% compared to the LR model, which is extensively usedin the previous studies for airfare price prediction.To show the importance of features specifically employedfor our regression model, another experiment was conducted.In this experiment, we only used common features withvery high importance scores such as “Distance”, “SeatClass”, and “Passenger Volume”. The results are presentedin Table VI. Again, we find that LR and SVM have lowerperformance compared to other models, and RF reaches the2highest performance with respect to both RMSE and Radj.2However, the performance (Radj ) dropped by almost 7% forthe RF model when the less important factors are removed.Similarly, the performance for other models dropped aswell. Although the less important factors may not contributesignificantly to the performance, these results show that toachieve the best performing model, one should include the“Load factor”, “Competition Factor”, “CPI”, and “Crude OilPrice” as features. Consequently, the proposed frameworkutilizes all of these features to achieve the highest airfareprice prediction performance.V. C ONCLUSION AND F UTURE W ORKIn this study, a machine learning framework was developed to predict the quarterly average airfare price onthe market segment level. We combined the U.S. domesticairline tickets sales data and non-stop segment data fromtwo public datasets (DB1B and T-100). Several featureswere extracted from the datasets and combined togetherwith macroeconomic data, to model the air travel marketsegments. With the help of the feature selection techniques,our proposed model is able to predict the quarterly averageairfare price with an adjusted R squared score of 0.869.To the best of our knowledge, most of previous studies onairfare price prediction using the DB1B dataset have focusedon conventional statistical approaches, which have their ownlimitations of problem estimations and assumptions. Also,to our knowledge, no other studies have integrated theinformation from DB1B, T-100, and m
Coupon DB1B market segment, time of the itinerary, carrier, and seat class Market segment DB1B&T-100 original airport, destination airport, and segment distance Market segment by carrier T-100 number of passengers, and number of available seats by aircraft type III. MATERIALS AND METHOD A. Datasets In ord
Bruksanvisning för bilstereo . Bruksanvisning for bilstereo . Instrukcja obsługi samochodowego odtwarzacza stereo . Operating Instructions for Car Stereo . 610-104 . SV . Bruksanvisning i original
10 tips och tricks för att lyckas med ert sap-projekt 20 SAPSANYTT 2/2015 De flesta projektledare känner säkert till Cobb’s paradox. Martin Cobb verkade som CIO för sekretariatet för Treasury Board of Canada 1995 då han ställde frågan
service i Norge och Finland drivs inom ramen för ett enskilt företag (NRK. 1 och Yleisradio), fin ns det i Sverige tre: Ett för tv (Sveriges Television , SVT ), ett för radio (Sveriges Radio , SR ) och ett för utbildnings program (Sveriges Utbildningsradio, UR, vilket till följd av sin begränsade storlek inte återfinns bland de 25 största
Hotell För hotell anges de tre klasserna A/B, C och D. Det betyder att den "normala" standarden C är acceptabel men att motiven för en högre standard är starka. Ljudklass C motsvarar de tidigare normkraven för hotell, ljudklass A/B motsvarar kraven för moderna hotell med hög standard och ljudklass D kan användas vid
LÄS NOGGRANT FÖLJANDE VILLKOR FÖR APPLE DEVELOPER PROGRAM LICENCE . Apple Developer Program License Agreement Syfte Du vill använda Apple-mjukvara (enligt definitionen nedan) för att utveckla en eller flera Applikationer (enligt definitionen nedan) för Apple-märkta produkter. . Applikationer som utvecklas för iOS-produkter, Apple .
Speake/s discretion, the departure/return city may vary for the first class airfare. International Events: Customer shall be responsible for roundtrip, first class commercial airfare for two between Anchorage, Alaska and event city and round-trip business class airfare for two between event city and the lower 48 US States. At the
och krav. Maskinerna skriver ut upp till fyra tum breda etiketter med direkt termoteknik och termotransferteknik och är lämpliga för en lång rad användningsområden på vertikala marknader. TD-seriens professionella etikettskrivare för . skrivbordet. Brothers nya avancerade 4-tums etikettskrivare för skrivbordet är effektiva och enkla att
Den kanadensiska språkvetaren Jim Cummins har visat i sin forskning från år 1979 att det kan ta 1 till 3 år för att lära sig ett vardagsspråk och mellan 5 till 7 år för att behärska ett akademiskt språk.4 Han införde två begrepp för att beskriva elevernas språkliga kompetens: BI
